Re: [PATCH v9 4/4] ree: Improve ree pass for rs6000 target using defined ABI interfaces

[Ajit Agarwal <aagarwa1@linux.ibm.com>]

Hello Vineet:

Thanks for your time and valuable comments.

On 21/10/23 5:26 am, Vineet Gupta wrote:
> On 10/19/23 23:50, Ajit Agarwal wrote:
>> Hello All:
>>
>> This version 9 of the patch uses abi interfaces to remove zero and sign extension elimination.
>> Bootstrapped and regtested on powerpc-linux-gnu.
>>
>> In this version (version 9) of the patch following review comments are incorporated.
>>
>> a) Removal of hard code zero_extend and sign_extend  in abi interfaces.
>> b) Source and destination with different registers are considered.
>> c) Further enhancements.
>> d) Added sign extension elimination using abi interfaces.
> 
> As has been trend in the past, I don't think all the review comments have been addressed.
> The standard practice is to reply to reviewer's email and say yay/nay explicitly to each comment. Some of my comments in [1a] are still not resolved, importantly the last two. To be fair you did reply [1b] but the comments were not addressed explicitly.
> 
> [1a] https://gcc.gnu.org/pipermail/gcc-patches/2023-September/630814.html
> [1b] https://gcc.gnu.org/pipermail/gcc-patches/2023-September/630865.html
> 

I have addressed the last 2 comments in the version 10 of the patch. Please let me know if there
is anything missing. Regarding last comments with a providing different tests if you have any suggestions please 
let me know.

> Anyhow I gave this a try for RISC-V, specially after [2a][2b] as I was curious to see if this uncovers REE handling extraneous extensions which could potentially be eliminated in Expand itself, which is generally better as it happens earlier in the pipeline.
> 
> [2a] 2023-10-16 8eb9cdd14218 expr: don't clear SUBREG_PROMOTED_VAR_P flag for a promoted subreg [target/111466]
> [2b] https://gcc.gnu.org/pipermail/gcc-patches/2023-October/631818.html
> 
> Bad news is with the patch, we fail to even bootstrap risc-v, buckles over when building libgcc itself.
> 
> The reproducer is embarrassingly simple, build with -O2:
> 
> float a;
> b() { return a; }
> 
> See details below....
> 
>> Thanks & Regards
>> Ajit
>>
>> ree: Improve ree pass for rs6000 target using defined abi interfaces
>>
>> For rs6000 target we see redundant zero and sign extension and done
>> to improve ree pass to eliminate such redundant zero and sign extension
>> using defined ABI interfaces.
>>
>> 2023-10-20  Ajit Kumar Agarwal  <aagarwa1@linux.ibm.com>
>>
>> gcc/ChangeLog:
>>
>>          * ree.cc (combine_reaching_defs): Use of zero_extend and sign_extend
>>          defined abi interfaces.
>>          (add_removable_extension): Use of defined abi interfaces for no
>>          reaching defs.
>>          (abi_extension_candidate_return_reg_p): New function.
>>          (abi_extension_candidate_p): New function.
>>          (abi_extension_candidate_argno_p): New function.
>>          (abi_handle_regs): New function.
>>          (abi_target_promote_function_mode): New function.
>>
>> gcc/testsuite/ChangeLog:
>>
>>          * g++.target/powerpc/zext-elim-3.C
>> ---
>>   gcc/ree.cc                                    | 151 +++++++++++++++++-
>>   .../g++.target/powerpc/zext-elim-3.C          |  13 ++
>>   2 files changed, 161 insertions(+), 3 deletions(-)
>>   create mode 100644 gcc/testsuite/g++.target/powerpc/zext-elim-3.C
>>
>> diff --git a/gcc/ree.cc b/gcc/ree.cc
>> index fc04249fa84..9f21f0e9907 100644
>> --- a/gcc/ree.cc
>> +++ b/gcc/ree.cc
>> @@ -514,7 +514,8 @@ get_uses (rtx_insn *insn, rtx reg)
>>       if (REGNO (DF_REF_REG (def)) == REGNO (reg))
>>         break;
>>   -  gcc_assert (def != NULL);
>> +  if (def == NULL)
>> +    return NULL;
>>       ref_chain = DF_REF_CHAIN (def);
>>   @@ -750,6 +751,124 @@ get_extended_src_reg (rtx src)
>>     return src;
>>   }
>>   +/* Return TRUE if target mode is equal to source mode of zero_extend
>> +   or sign_extend otherwise false.  */
>> +
>> +static bool
>> +abi_target_promote_function_mode (machine_mode mode)
>> +{
>> +  int unsignedp;
>> +  machine_mode tgt_mode
>> +    = targetm.calls.promote_function_mode (NULL_TREE, mode, &unsignedp,
>> +                       NULL_TREE, 1);
>> +
>> +  if (tgt_mode == mode)
>> +    return true;
>> +  else
>> +    return false;
>> +}
>> +
>> +/* Return TRUE if the candidate insn is zero extend and regno is
>> +   a return registers.  */
>> +
>> +static bool
>> +abi_extension_candidate_return_reg_p (/*rtx_insn *insn, */int regno)
>> +{
>> +  if (targetm.calls.function_value_regno_p (regno))
>> +    return true;
>> +
>> +  return false;
>> +}
>> +
>> +/* Return TRUE if reg source operand of zero_extend is argument registers
>> +   and not return registers and source and destination operand are same
>> +   and mode of source and destination operand are not same.  */
>> +
>> +static bool
>> +abi_extension_candidate_p (rtx_insn *insn)
>> +{
>> +  rtx set = single_set (insn);
>> +  machine_mode dst_mode = GET_MODE (SET_DEST (set));
>> +  rtx orig_src = XEXP (SET_SRC (set), 0);
>> +
>> +  if (!FUNCTION_ARG_REGNO_P (REGNO (orig_src))
>> +      || abi_extension_candidate_return_reg_p (/*insn,*/ REGNO (orig_src)))
>> +    return false;
>> +
>> +  /* Mode of destination and source should be different.  */
>> +  if (dst_mode == GET_MODE (orig_src))
>> +    return false;
>> +
>> +  machine_mode mode = GET_MODE (XEXP (SET_SRC (set), 0));
>> +  bool promote_p = abi_target_promote_function_mode (mode);
>> +
>> +  /* REGNO of source and destination should be same if not
>> +      promoted.  */
>> +  if (!promote_p && REGNO (SET_DEST (set)) != REGNO (orig_src))
>> +    return false;
>> +
>> +  return true;
>> +}
>> +
>> +/* Return TRUE if the candidate insn is zero extend and regno is
>> +   an argument registers.  */
>> +
>> +static bool
>> +abi_extension_candidate_argno_p (/*rtx_code code, */int regno)
>> +{
>> +  if (FUNCTION_ARG_REGNO_P (regno))
>> +    return true;
>> +
>> +  return false;
>> +}
>> +
>> +/* Return TRUE if the candidate insn doesn't have defs and have
>> + * uses without RTX_BIN_ARITH/RTX_COMM_ARITH/RTX_UNARY rtx class.  */
>> +
>> +static bool
>> +abi_handle_regs (rtx_insn *insn)
>> +{
>> +  if (side_effects_p (PATTERN (insn)))
>> +    return false;
>> +
>> +  struct df_link *uses = get_uses (insn, SET_DEST (PATTERN (insn)));
>> +
>> +  if (!uses)
>> +    return false;
>> +
>> +  for (df_link *use = uses; use; use = use->next)
>> +    {
>> +      if (!use->ref)
>> +    return false;
>> +
>> +      if (BLOCK_FOR_INSN (insn) != BLOCK_FOR_INSN (DF_REF_INSN (use->ref)))
>> +    return false;
>> +
>> +      rtx_insn *use_insn = DF_REF_INSN (use->ref);
>> +
>> +      if (GET_CODE (PATTERN (use_insn)) == SET)
>> +    {
>> +      rtx_code code = GET_CODE (SET_SRC (PATTERN (use_insn)));
>> +
>> +      if (GET_RTX_CLASS (code) == RTX_BIN_ARITH
>> +          || GET_RTX_CLASS (code) == RTX_COMM_ARITH
>> +          || GET_RTX_CLASS (code) == RTX_UNARY)
>> +        return false;
>> +    }
>> +     }
>> +
>> +  rtx set = single_set (insn);
>> +
>> +  if (GET_CODE (SET_SRC (set)) == SIGN_EXTEND)
>> +    {
>> +      machine_mode mode = GET_MODE (XEXP (SET_SRC (set), 0));
>> +      bool promote_p = abi_target_promote_function_mode (mode);
>> +
>> +      return promote_p;
>> +    }
>> +  return true;
>> +}
>> +
>>   /* This function goes through all reaching defs of the source
>>      of the candidate for elimination (CAND) and tries to combine
>>      the extension with the definition instruction.  The changes
>> @@ -770,6 +889,11 @@ combine_reaching_defs (ext_cand *cand, const_rtx set_pat, ext_state *state)
>>       state->defs_list.truncate (0);
>>     state->copies_list.truncate (0);
>> +  rtx orig_src = XEXP (SET_SRC (cand->expr),0);
>> +
>> +  if (abi_extension_candidate_p (cand->insn)
>> +      && (!get_defs (cand->insn, orig_src, NULL)))
>> +    return abi_handle_regs (cand->insn);
>>       outcome = make_defs_and_copies_lists (cand->insn, set_pat, state);
>>   @@ -1036,6 +1160,15 @@ combine_reaching_defs (ext_cand *cand, const_rtx set_pat, ext_state *state)
>>           }
>>       }
>>   +  rtx insn_set = single_set (cand->insn);
>> +
>> +  machine_mode mode = (GET_MODE (XEXP (SET_SRC (insn_set), 0)));
>> +
>> +  bool promote_p = abi_target_promote_function_mode (mode);
>> +
>> +  if (promote_p)
>> +    return true;
>> +
> 
> Is this early exit OK ? It skips a subsequent apply_change_group() call which could potentially fail and thus extension would be undone.
> And even for the passing case, we do want those instructions to be merged normally. So I don't understand how this change is useful at all ?
> 
> FWIW for my test, w/o the patch: apply_change_group would fail (likely missing some combine pattern) and undo the merge. However w/ patch we just return/continue, keeping the merged but invalid insn which ICE in next pass cprop_hardreg as failure to recog that insn.
> 
> Some more details in case you are curious:
> 
> Coming into REE we have
> 
> (insn 7 6 13 2 (set (reg:SI 10 a0 [136])  #DEF
>         (fix:SI (reg:SF 47 fa5 [orig:138 a ] [138])))
>                  {fix_truncsfsi2}
> 
> (insn 13 7 14 2 (set (reg/i:DI 10 a0)    # extension
>         (sign_extend:DI (reg:SI 10 a0 [136])))
>                  {extendsidi2}
> 
> These are merged into unrecog insn as of now:
> 
> (insn 7 6 14 2 (set (reg:DI 10 a0)
>         (sign_extend:DI (fix:SI (reg:SF 47 fa5 [orig:138 a ] [138]))))
>

Thanks for your input on this. I have addressed that in the version 10 of the patch please review.

Thanks & Regards
Ajit
 
> Thx,
> -Vineet
> 
>>     if (merge_successful)
>>       {
>>         /* Commit the changes here if possible
>> @@ -1112,6 +1245,14 @@ add_removable_extension (const_rtx expr, rtx_insn *insn,
>>         rtx reg = XEXP (src, 0);
>>         struct df_link *defs, *def;
>>         ext_cand *cand;
>> +      defs = get_defs (insn, reg, NULL);
>> +
>> +      if (!defs && abi_extension_candidate_argno_p (/*code,*/ REGNO (reg)))
>> +    {
>> +      ext_cand e = {expr, code, mode, insn};
>> +      insn_list->safe_push (e);
>> +      return;
>> +    }
>>           /* Zero-extension of an undefined value is partly defined (it's
>>        completely undefined for sign-extension, though).  So if there exists
>> @@ -1131,7 +1272,6 @@ add_removable_extension (const_rtx expr, rtx_insn *insn,
>>       }
>>           /* Second, make sure we can get all the reaching definitions.  */
>> -      defs = get_defs (insn, reg, NULL);
>>         if (!defs)
>>       {
>>         if (dump_file)
>> @@ -1321,7 +1461,8 @@ find_and_remove_re (void)
>>             && (REGNO (SET_DEST (set)) != REGNO (XEXP (SET_SRC (set), 0))))
>>           {
>>                 reinsn_copy_list.safe_push (curr_cand->insn);
>> -              reinsn_copy_list.safe_push (state.defs_list[0]);
>> +          if (state.defs_list.length () != 0)
>> +        reinsn_copy_list.safe_push (state.defs_list[0]);
>>           }
>>         reinsn_del_list.safe_push (curr_cand->insn);
>>         state.modified[INSN_UID (curr_cand->insn)].deleted = 1;
>> @@ -1345,6 +1486,10 @@ find_and_remove_re (void)
>>     for (unsigned int i = 0; i < reinsn_copy_list.length (); i += 2)
>>       {
>>         rtx_insn *curr_insn = reinsn_copy_list[i];
>> +
>> +      if ((i+1) >= reinsn_copy_list.length ())
>> +    continue;
>> +
>>         rtx_insn *def_insn = reinsn_copy_list[i + 1];
>>           /* Use the mode of the destination of the defining insn
>> diff --git a/gcc/testsuite/g++.target/powerpc/zext-elim-3.C b/gcc/testsuite/g++.target/powerpc/zext-elim-3.C
>> new file mode 100644
>> index 00000000000..5a050df06ff
>> --- /dev/null
>> +++ b/gcc/testsuite/g++.target/powerpc/zext-elim-3.C
>> @@ -0,0 +1,13 @@
>> +/* { dg-options "-mcpu=power9 -O2" } */
>> +
>> +void *memset(void *b, int c, unsigned long len)
>> +{
>> +  unsigned long i;
>> +
>> +  for (i = 0; i < len; i++)
>> +    ((unsigned char *)b)[i] = c;
>> +
>> +   return b;
>> +}
>> +
>> +/* { dg-final { scan-assembler-not "\mrlwinm\M" } } */
>