diff mbox series

[aarch64] v4: Preparatory patch to place target independent and,dependent changed code in one file

Message ID 3f463631-216a-486f-9907-9da7848f22e0@linux.ibm.com
State New
Headers show
Series [aarch64] v4: Preparatory patch to place target independent and,dependent changed code in one file | expand

Commit Message

Ajit Agarwal May 14, 2024, 9:38 a.m. UTC
Hello Alex/Richard:

All comments are addressed.

Common infrastructure of load store pair fusion is divided into target
independent and target dependent changed code.

Target independent code is the Generic code with pure virtual function
to interface betwwen target independent and dependent code.

Target dependent code is the implementation of pure virtual function for
aarch64 target and the call to target independent code.

Bootstrapped on aarch64-linux-gnu.

Thanks & Regards
Ajit



arch64: Preparatory patch to place target independent and
dependent changed code in one file

Common infrastructure of load store pair fusion is divided into target
independent and target dependent changed code.

Target independent code is the Generic code with pure virtual function
to interface betwwen target independent and dependent code.

Target dependent code is the implementation of pure virtual function for
aarch64 target and the call to target independent code.

2024-05-14  Ajit Kumar Agarwal  <aagarwa1@linux.ibm.com>

gcc/ChangeLog:

	* config/aarch64/aarch64-ldp-fusion.cc: Place target
	independent and dependent changed code.
---
 gcc/config/aarch64/aarch64-ldp-fusion.cc | 526 +++++++++++++++--------
 1 file changed, 345 insertions(+), 181 deletions(-)

Comments

Alex Coplan May 14, 2024, 6:23 p.m. UTC | #1
Hi Ajit,

Please can you pay careful attention to the review comments?

In particular, you have ignored my comment about changing the access of
member functions in ldp_bb_info several times now (on at least three
patch reviews).

Likewise on multiple occasions you've only partially implemented a piece
of review feedback (e.g. applying the "override" keyword to virtual
overrides).

That all makes it rather tiresome to review your patches.

Also, I realise I should have mentioned this on a previous revision of
this patch, but I thought we previously agreed (with Richard S) to split
out the renaming in existing code (e.g. ldp/stp -> "paired access" and
so on) to a separate patch?  That would make this eaiser to review.

On 14/05/2024 15:08, Ajit Agarwal wrote:
> Hello Alex/Richard:
> 
> All comments are addressed.
> 
> Common infrastructure of load store pair fusion is divided into target
> independent and target dependent changed code.
> 
> Target independent code is the Generic code with pure virtual function
> to interface betwwen target independent and dependent code.
> 
> Target dependent code is the implementation of pure virtual function for
> aarch64 target and the call to target independent code.
> 
> Bootstrapped on aarch64-linux-gnu.
> 
> Thanks & Regards
> Ajit
> 
> 
> 
> arch64: Preparatory patch to place target independent and
> dependent changed code in one file
> 
> Common infrastructure of load store pair fusion is divided into target
> independent and target dependent changed code.
> 
> Target independent code is the Generic code with pure virtual function
> to interface betwwen target independent and dependent code.
> 
> Target dependent code is the implementation of pure virtual function for
> aarch64 target and the call to target independent code.
> 
> 2024-05-14  Ajit Kumar Agarwal  <aagarwa1@linux.ibm.com>
> 
> gcc/ChangeLog:
> 
> 	* config/aarch64/aarch64-ldp-fusion.cc: Place target
> 	independent and dependent changed code.
> ---
>  gcc/config/aarch64/aarch64-ldp-fusion.cc | 526 +++++++++++++++--------
>  1 file changed, 345 insertions(+), 181 deletions(-)
> 
> diff --git a/gcc/config/aarch64/aarch64-ldp-fusion.cc b/gcc/config/aarch64/aarch64-ldp-fusion.cc
> index 1d9caeab05d..e6af4b0570a 100644
> --- a/gcc/config/aarch64/aarch64-ldp-fusion.cc
> +++ b/gcc/config/aarch64/aarch64-ldp-fusion.cc
> @@ -138,6 +138,210 @@ struct alt_base
>    poly_int64 offset;
>  };
>  
> +// Virtual base class for load/store walkers used in alias analysis.
> +struct alias_walker
> +{
> +  virtual bool conflict_p (int &budget) const = 0;
> +  virtual insn_info *insn () const = 0;
> +  virtual bool valid () const = 0;
> +  virtual void advance () = 0;
> +};
> +
> +// This is used in handle_writeback_opportunities describing
> +// ALL if aarch64_ldp_writeback > 1 otherwise check
> +// EXISTING if aarch64_ldp_writeback.

Since this enum belongs to the generic interface, it's best if it is
described in general terms, i.e. the comment shouldn't refer to the
aarch64 param.

How about:

// When querying handle_writeback_opportunities, this enum is used to
// qualify which opportunities we are asking about.

then above the EXISTING enumerator, you could say:

  // Only those writeback opportunities that arise from existing
  // auto-increment accesses.

and for ALL, you could say:

  // All writeback opportunities including those that involve folding
  // base register updates into a non-writeback pair.

> +enum class writeback {
> +  ALL,
> +  EXISTING
> +};

Also, sorry for the very minor nit, but I think it is more logical if we
flip the order of the enumerators here, i.e. EXISTING should come first.

> +
> +struct pair_fusion {
> +  pair_fusion ()
> +  {
> +    calculate_dominance_info (CDI_DOMINATORS);
> +    df_analyze ();
> +    crtl->ssa = new rtl_ssa::function_info (cfun);
> +  };
> +
> +  // Given:
> +  // - an rtx REG_OP, the non-memory operand in a load/store insn,
> +  // - a machine_mode MEM_MODE, the mode of the MEM in that insn, and
> +  // - a boolean LOAD_P (true iff the insn is a load), then:
> +  // return true if the access should be considered an FP/SIMD access.
> +  // Such accesses are segregated from GPR accesses, since we only want
> +  // to form pairs for accesses that use the same register file.
> +  virtual bool fpsimd_op_p (rtx, machine_mode, bool)
> +  {
> +    return false;
> +  }
> +
> +  // Return true if we should consider forming ldp/stp insns from memory

Replace "ldp/stp insns" with "pairs" here, since this is the generic
interface.

> +  // accesses with operand mode MODE at this stage in compilation.
> +  virtual bool pair_operand_mode_ok_p (machine_mode mode) = 0;
> +
> +  // Return true iff REG_OP is a suitable register operand for a paired
> +  // memory access, where LOAD_P is true if we're asking about loads and
> +  // false for stores.  MODE gives the mode of the operand.
> +  virtual bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
> +				      machine_mode mode) = 0;
> +
> +  // Return alias check limit.
> +  // This is needed to avoid unbounded quadratic behaviour when
> +  // performing alias analysis.
> +  virtual int pair_mem_alias_check_limit () = 0;
> +
> +  // Returns true if we should try to handle writeback opportunities

Missing punctuation at the end of the sentence here.

> +  // WHICH parameter decides ALL or EXISTING writeback pairs.

How about: "WHICH determines the kinds of writeback opportunities the
caller is asking about."

> +  virtual bool handle_writeback_opportunities (enum writeback which) = 0 ;
> +
> +  // Given BASE_MEM, the mem from the lower candidate access for a pair,
> +  // and LOAD_P (true if the access is a load), check if we should proceed
> +  // to form the pair given the target's code generation policy on
> +  // paired accesses.
> +  virtual bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) = 0;
> +
> +  // Generate the pattern for a paired access. PATS gives the patterns

Nit: two spaces after '.'.  Run your patch through
contrib/check_GNU_style.py before submitting, please.

> +  // for the individual memory accesses (which by this point must share a
> +  // common base register).  If WRITEBACK is non-NULL, then this rtx
> +  // describes the update to the base register that should be performed by
> +  // the resulting insn.  LOAD_P is true iff the accesses are loads.
> +  virtual rtx gen_pair (rtx *pats, rtx writeback, bool load_p) = 0;
> +
> +  // Return true if memory is paired access, given INSN and LOAD_P
> +  // is true for load insn and false for store insn.
> +  virtual bool pair_mem_insn_p (rtx_insn *, bool &) = 0;

Since this is now a pure virtual, you can use parameter names in the
prototype, then refer to them directly in the comment, so something
like:

// Return true if INSN is a paired memory access.  If so, set LOAD_P to
// true iff INSN is a load pair.

> +
> +  // Return true if we should track loads.
> +  virtual bool track_loads_p ()
> +  {
> +    return true;
> +  }
> +
> +  // Return true if we should track stores.
> +  virtual bool track_stores_p ()
> +  {
> +    return true;
> +  }
> +
> +  // Return true if OFF  is in range.

Excess space after OFF.  Sorry for missing this last time, but "in
range" by itself isn't meaningful, how about "in range for a paired
memory access" instead.

> +  virtual bool pair_mem_in_range_p (HOST_WIDE_INT off) = 0;

Sorry for the nit, but I suggested last time renaming s/off/offset/.
Let's do that, please.

> +
> +  // Given a load/store pair insn in PATTERN, unpack the insn, storing
> +  // the register operands in REGS, and returning the mem. LOAD_P is true

Nit: two spaces after '.'.

> +  // for load and false for store.

Sorry for the nit, but I think the last sentence should say "LOAD_P is
true for loads and false for stores." (i.e. plural loads/stores).

> +  virtual rtx destructure_pair (rtx regs[2], rtx pattern, bool load_p) = 0;
> +
> +  // Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx

The name PAIR_MEM doesn't match the parameter name.  One of them needs
to change.

> +  // representing the effect of writeback on the base register in WB_EFFECT,
> +  // return an insn representing a writeback variant of this pair.
> +  // LOAD_P is true iff the pair is a load.
> +  // This is used when promoting existing non-writeback pairs to writeback
> +  // variants.
> +  virtual rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem,
> +					  rtx regs[2], bool load_p) = 0;
> +
> +  void ldp_fusion_bb (bb_info *bb);
> +  inline insn_info *find_trailing_add (insn_info *insns[2],
> +				       const insn_range_info &pair_range,
> +				       int initial_writeback,
> +				       rtx *writeback_effect,
> +				       def_info **add_def,
> +				       def_info *base_def,
> +				       poly_int64 initial_offset,
> +				       unsigned access_size);
> +  inline int get_viable_bases (insn_info *insns[2],
> +			       vec<base_cand> &base_cands,
> +			       rtx cand_mems[2],
> +			       unsigned access_size,
> +			       bool reversed);
> +  inline void do_alias_analysis (insn_info *alias_hazards[4],
> +				 alias_walker *walkers[4],
> +				 bool load_p);
> +  inline void try_promote_writeback (insn_info *insn, bool load_p);
> +  inline void run ();
> +  ~pair_fusion()

Nit: space before '('.  Again, contrib/check_GNU_style.py can help here.

> +  {
> +    if (crtl->ssa->perform_pending_updates ())
> +      cleanup_cfg (0);
> +
> +    free_dominance_info (CDI_DOMINATORS);
> +
> +    delete crtl->ssa;
> +    crtl->ssa = nullptr;
> +  }
> +};
> +
> +// This is the main function to start the pass.
> +void
> +pair_fusion::run ()
> +{
> +  if (!track_loads_p () && !track_stores_p ())
> +    return;
> +
> +  for (auto bb : crtl->ssa->bbs ())
> +    ldp_fusion_bb (bb);
> +}
> +
> +struct aarch64_pair_fusion : public pair_fusion
> +{
> +  // Before RA, we use the modes, noting that stores of constant zero
> +  // operands use GPRs (even in non-integer modes).  After RA, we use
> +  // the hard register numbers.

Sorry for the very minor nit, but I think this comment would fit better
inside the function body (above the return statement).

> +  bool fpsimd_op_p (rtx reg_op, machine_mode mem_mode,
> +		    bool load_p) override final
> +  {
> +    return reload_completed
> +      ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
> +      : (GET_MODE_CLASS (mem_mode) != MODE_INT
> +	 && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
> +  }
> +
> +  bool pair_mem_insn_p (rtx_insn *rti, bool &load_p) override final;
> +  bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) override final
> +  {
> +    return aarch64_mem_ok_with_ldpstp_policy_model (base_mem,
> +						    load_p,
> +						    GET_MODE (base_mem));
> +  }

Can we add blank lines between all of the inline member function
implementations in this class, please?  I did ask about that last time.

> +  bool pair_operand_mode_ok_p (machine_mode mode);

Looks like you're still missing "override final" on all of these
remaining overrides.

> +  rtx gen_pair (rtx *pats, rtx writeback, bool load_p);
> +  bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
> +			      machine_mode mode)
> +  {
> +    return (load_p
> +	    ? aarch64_ldp_reg_operand (reg_op, mode)
> +	    : aarch64_stp_reg_operand (reg_op, mode));
> +  }
> +  int pair_mem_alias_check_limit ()
> +  {
> +    return aarch64_ldp_alias_check_limit;
> +  }
> +  bool handle_writeback_opportunities (enum writeback which)
> +  {
> +    if (which == writeback::ALL)
> +      return aarch64_ldp_writeback > 1;
> +    else
> +      return aarch64_ldp_writeback;
> +  }
> +  bool track_loads_p ()
> +  {
> +    return aarch64_tune_params.ldp_policy_model
> +	   != AARCH64_LDP_STP_POLICY_NEVER;
> +  }
> +  bool track_stores_p ()
> +  {
> +    return aarch64_tune_params.stp_policy_model
> +	   != AARCH64_LDP_STP_POLICY_NEVER;
> +  }
> +  bool pair_mem_in_range_p (HOST_WIDE_INT off)
> +  {
> +    return (off >= LDP_MIN_IMM && off <= LDP_MAX_IMM);
> +  }
> +  rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem, rtx regs[2],
> +				  bool load_p);
> +  rtx destructure_pair (rtx regs[2], rtx rti, bool load_p);
> +};
> +
>  // State used by the pass for a given basic block.
>  struct ldp_bb_info
>  {
> @@ -159,9 +363,9 @@ struct ldp_bb_info
>    hash_map<def_hash, alt_base> canon_base_map;
>  
>    static const size_t obstack_alignment = sizeof (void *);
> -  bb_info *m_bb;
>  
> -  ldp_bb_info (bb_info *bb) : m_bb (bb), m_emitted_tombstone (false)
> +  ldp_bb_info (bb_info *bb, pair_fusion *d)
> +	       : m_bb (bb), m_pass (d), m_emitted_tombstone (false)
>    {
>      obstack_specify_allocation (&m_obstack, OBSTACK_CHUNK_SIZE,
>  				obstack_alignment, obstack_chunk_alloc,
> @@ -181,37 +385,75 @@ struct ldp_bb_info
>    inline void track_access (insn_info *, bool load, rtx mem);
>    inline void transform ();
>    inline void cleanup_tombstones ();
> +  inline void merge_pairs (insn_list_t &, insn_list_t &,
> +		    bool load_p, unsigned access_size);
> +  inline void transform_for_base (int load_size, access_group &group);
> +
> +  inline bool try_fuse_pair (bool load_p, unsigned access_size,
> +			     insn_info *i1, insn_info *i2);
> +
> +  inline bool fuse_pair (bool load_p, unsigned access_size,
> +		  int writeback,
> +		  insn_info *i1, insn_info *i2,
> +		  base_cand &base,
> +		  const insn_range_info &move_range);
> +
> +  inline void track_tombstone (int uid);

You keep missing my review comments here.  Why are you changing these
member functions from private to public?

>  
>  private:
>    obstack m_obstack;
> -

Can we keep the blank lines here?  They were there to logically group
together the members related to tracking tombstones.

> +  bb_info *m_bb;
> +  pair_fusion *m_pass;
>    // State for keeping track of tombstone insns emitted for this BB.
>    bitmap_obstack m_bitmap_obstack;
>    bitmap_head m_tombstone_bitmap;
>    bool m_emitted_tombstone;
> -
> +  template<typename Map> inline void traverse_base_map (Map &map);
> +  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
>    inline splay_tree_node<access_record *> *node_alloc (access_record *);
> +};

What is the purpose of these changes?  Why are you moving these
function declarations around?

>  
> -  template<typename Map>
> -  inline void traverse_base_map (Map &map);
> -  inline void transform_for_base (int load_size, access_group &group);
> -
> -  inline void merge_pairs (insn_list_t &, insn_list_t &,
> -			   bool load_p, unsigned access_size);
> -
> -  inline bool try_fuse_pair (bool load_p, unsigned access_size,
> -			     insn_info *i1, insn_info *i2);
> +bool
> +aarch64_pair_fusion::pair_mem_insn_p (rtx_insn *rti,
> +				      bool &load_p)

No need for the line break here, the params should fit on a single line.

> +{
> +  rtx pat = PATTERN (rti);
> +  if (GET_CODE (pat) == PARALLEL
> +      && XVECLEN (pat, 0) == 2)
> +    {
> +      const auto attr = get_attr_ldpstp (rti);
> +      if (attr == LDPSTP_NONE)
> +	return false;
>  
> -  inline bool fuse_pair (bool load_p, unsigned access_size,
> -			 int writeback,
> -			 insn_info *i1, insn_info *i2,
> -			 base_cand &base,
> -			 const insn_range_info &move_range);
> +      load_p = (attr == LDPSTP_LDP);
> +      gcc_checking_assert (load_p || attr == LDPSTP_STP);
> +      return true;
> +    }
> +  return false;
> +}
>  
> -  inline void track_tombstone (int uid);
> +rtx
> +aarch64_pair_fusion::gen_pair (rtx *pats,
> +			       rtx writeback,
> +			       bool load_p)

Likewise, the params should all fit on a single line.

> +{
> +  rtx pair_pat;
>  
> -  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
> -};
> +  if (writeback)
> +    {
> +      auto patvec = gen_rtvec (3, writeback, pats[0], pats[1]);
> +      return  gen_rtx_PARALLEL (VOIDmode, patvec);

Excess space after return.

> +    }
> +  else if (load_p)
> +    return aarch64_gen_load_pair (XEXP (pats[0], 0),
> +				  XEXP (pats[1], 0),
> +				  XEXP (pats[0], 1));
> +  else
> +    return aarch64_gen_store_pair (XEXP (pats[0], 0),
> +				   XEXP (pats[0], 1),
> +				   XEXP (pats[1], 1));
> +  return pair_pat;
> +}
>  
>  splay_tree_node<access_record *> *
>  ldp_bb_info::node_alloc (access_record *access)
> @@ -312,8 +554,8 @@ any_post_modify_p (rtx x)
>  
>  // Return true if we should consider forming ldp/stp insns from memory
>  // accesses with operand mode MODE at this stage in compilation.
> -static bool
> -ldp_operand_mode_ok_p (machine_mode mode)
> +bool
> +aarch64_pair_fusion::pair_operand_mode_ok_p (machine_mode mode)
>  {
>    if (!aarch64_ldpstp_operand_mode_p (mode))
>      return false;
> @@ -404,9 +646,10 @@ ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
>    const machine_mode mem_mode = GET_MODE (mem);
>    const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
>  
> -  // Punt on misaligned offsets.  LDP/STP instructions require offsets to be a
> -  // multiple of the access size, and we believe that misaligned offsets on
> -  // MEM_EXPR bases are likely to lead to misaligned offsets w.r.t. RTL bases.
> +  // Punt on misaligned offsets.  Paired memory access instructions require
> +  // offsets to be a multiple of the access size, and we believe that
> +  // misaligned offsets on MEM_EXPR bases are likely to lead to misaligned
> +  // offsets w.r.t. RTL bases.
>    if (!multiple_p (offset, mem_size))
>      return false;
>  
> @@ -430,10 +673,10 @@ ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
>  }
>  
>  // Main function to begin pair discovery.  Given a memory access INSN,
> -// determine whether it could be a candidate for fusing into an ldp/stp,
> -// and if so, track it in the appropriate data structure for this basic
> -// block.  LOAD_P is true if the access is a load, and MEM is the mem
> -// rtx that occurs in INSN.
> +// determine whether it could be a candidate for fusing into a paired
> +// access, and if so, track it in the appropriate data structure for
> +// this basic block.  LOAD_P is true if the access is a load, and MEM
> +// is the mem rtx that occurs in INSN.
>  void
>  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>  {
> @@ -441,35 +684,24 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>    if (MEM_VOLATILE_P (mem))
>      return;
>  
> -  // Ignore writeback accesses if the param says to do so.
> -  if (!aarch64_ldp_writeback
> +  // Ignore writeback accesses if the hook says to do so.
> +  if (!m_pass->handle_writeback_opportunities (writeback::EXISTING)
>        && GET_RTX_CLASS (GET_CODE (XEXP (mem, 0))) == RTX_AUTOINC)
>      return;
>  
>    const machine_mode mem_mode = GET_MODE (mem);
> -  if (!ldp_operand_mode_ok_p (mem_mode))
> +  if (!m_pass->pair_operand_mode_ok_p (mem_mode))
>      return;
>  
>    rtx reg_op = XEXP (PATTERN (insn->rtl ()), !load_p);
>  
> -  // Ignore the access if the register operand isn't suitable for ldp/stp.
> -  if (load_p
> -      ? !aarch64_ldp_reg_operand (reg_op, mem_mode)
> -      : !aarch64_stp_reg_operand (reg_op, mem_mode))
> +  if (!m_pass->pair_reg_operand_ok_p (load_p, reg_op, mem_mode))
>      return;
>  
>    // We want to segregate FP/SIMD accesses from GPR accesses.
> -  //
> -  // Before RA, we use the modes, noting that stores of constant zero
> -  // operands use GPRs (even in non-integer modes).  After RA, we use
> -  // the hard register numbers.
> -  const bool fpsimd_op_p
> -    = reload_completed
> -    ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
> -    : (GET_MODE_CLASS (mem_mode) != MODE_INT
> -       && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
> +  const bool fpsimd_op_p = m_pass->fpsimd_op_p (reg_op, mem_mode, load_p);
>  
> -  // Note ldp_operand_mode_ok_p already rejected VL modes.
> +  // Note pair_operand_mode_ok_p already rejected VL modes.
>    const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
>    const lfs_fields lfs = { load_p, fpsimd_op_p, mem_size };
>  
> @@ -498,8 +730,8 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>    // elimination offset pre-RA, we should postpone forming pairs on such
>    // accesses until after RA.
>    //
> -  // As it stands, addresses with offsets in range for LDR but not
> -  // in range for LDP/STP are currently reloaded inefficiently,
> +  // As it stands, addresses in range for an individual load/store but not
> +  // for a paired access are currently reloaded inefficiently,
>    // ending up with a separate base register for each pair.
>    //
>    // In theory LRA should make use of
> @@ -510,11 +742,11 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>    // constraint; this early return means we never get to the code
>    // that calls targetm.legitimize_address_displacement.
>    //
> -  // So for now, it's better to punt when we can't be sure that the
> -  // offset is in range for LDP/STP.  Out-of-range cases can then be
> -  // handled after RA by the out-of-range LDP/STP peepholes.  Eventually, it
> -  // would be nice to handle known out-of-range opportunities in the
> -  // pass itself (for stack accesses, this would be in the post-RA pass).
> +  // It's better to punt when we can't be sure that the
> +  // offset is in range for paired access.  On aarch64, Out-of-range cases
> +  // can then be handled after RA by the out-of-range PAIR MEM  peepholes.

No need to change "LDP/STP" to "PAIR MEM " here, now that you've
qualified that sentence as being aarch64-specific.

> +  // Eventually, it would be nice to handle known out-of-range opportunities
> +  // in the pass itself (for stack accesses, this would be in the post-RA pass).
>    if (!reload_completed
>        && (REGNO (base) == FRAME_POINTER_REGNUM
>  	  || REGNO (base) == ARG_POINTER_REGNUM))
> @@ -565,8 +797,8 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>  	gcc_unreachable (); // Base defs should be unique.
>      }
>  
> -  // Punt on misaligned offsets.  LDP/STP require offsets to be a multiple of
> -  // the access size.
> +  // Punt on misaligned offsets.  Paired memory accesses require offsets
> +  // to be a multiple of the access size.
>    if (!multiple_p (mem_off, mem_size))
>      return;
>  
> @@ -1199,8 +1431,8 @@ extract_writebacks (bool load_p, rtx pats[2], int changed)
>  // base register.  If there is one, we choose the first such update after
>  // PAIR_DST that is still in the same BB as our pair.  We return the new def in
>  // *ADD_DEF and the resulting writeback effect in *WRITEBACK_EFFECT.
> -static insn_info *
> -find_trailing_add (insn_info *insns[2],
> +insn_info *
> +pair_fusion::find_trailing_add (insn_info *insns[2],
>  		   const insn_range_info &pair_range,
>  		   int initial_writeback,
>  		   rtx *writeback_effect,
> @@ -1278,7 +1510,7 @@ find_trailing_add (insn_info *insns[2],
>  
>    off_hwi /= access_size;
>  
> -  if (off_hwi < LDP_MIN_IMM || off_hwi > LDP_MAX_IMM)
> +  if (!pair_mem_in_range_p (off_hwi))
>      return nullptr;
>  
>    auto dump_prefix = [&]()
> @@ -1792,7 +2024,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>      {
>        if (dump_file)
>  	fprintf (dump_file,
> -		 "  ldp: i%d has wb but subsequent i%d has non-wb "
> +		 "  load pair: i%d has wb but subsequent i%d has non-wb "
>  		 "update of base (r%d), dropping wb\n",
>  		 insns[0]->uid (), insns[1]->uid (), base_regno);
>        gcc_assert (writeback_effect);
> @@ -1815,7 +2047,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>      }
>  
>    // If either of the original insns had writeback, but the resulting pair insn
> -  // does not (can happen e.g. in the ldp edge case above, or if the writeback
> +  // does not (can happen e.g. in the load pair edge case above, or if the writeback

Nit: long line here (comment needs re-wrapping).

>    // effects cancel out), then drop the def(s) of the base register as
>    // appropriate.
>    //
> @@ -1834,7 +2066,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>    // update of the base register and try and fold it in to make this into a
>    // writeback pair.
>    insn_info *trailing_add = nullptr;
> -  if (aarch64_ldp_writeback > 1
> +  if (m_pass->handle_writeback_opportunities (writeback::ALL)
>        && !writeback_effect
>        && (!load_p || (!refers_to_regno_p (base_regno, base_regno + 1,
>  					 XEXP (pats[0], 0), nullptr)
> @@ -1842,7 +2074,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>  					     XEXP (pats[1], 0), nullptr))))
>      {
>        def_info *add_def;
> -      trailing_add = find_trailing_add (insns, move_range, writeback,
> +      trailing_add = m_pass->find_trailing_add (insns, move_range, writeback,
>  					&writeback_effect,
>  					&add_def, base.def, offsets[0],
>  					access_size);

Looks like the parameters need re-indenting here.

> @@ -1855,35 +2087,19 @@ ldp_bb_info::fuse_pair (bool load_p,
>      }
>  
>    // Now that we know what base mem we're going to use, check if it's OK
> -  // with the ldp/stp policy.
> +  // with the pair mem policy.
>    rtx first_mem = XEXP (pats[0], load_p);
> -  if (!aarch64_mem_ok_with_ldpstp_policy_model (first_mem,
> -						load_p,
> -						GET_MODE (first_mem)))
> +  if (!m_pass->pair_mem_ok_with_policy (first_mem, load_p))
>      {
>        if (dump_file)
> -	fprintf (dump_file, "punting on pair (%d,%d), ldp/stp policy says no\n",
> +	fprintf (dump_file, "punting on pair (%d,%d), pair mem policy says no\n",
>  		 i1->uid (), i2->uid ());
>        return false;
>      }
>  
>    rtx reg_notes = combine_reg_notes (first, second, load_p);
>  
> -  rtx pair_pat;
> -  if (writeback_effect)
> -    {
> -      auto patvec = gen_rtvec (3, writeback_effect, pats[0], pats[1]);
> -      pair_pat = gen_rtx_PARALLEL (VOIDmode, patvec);
> -    }
> -  else if (load_p)
> -    pair_pat = aarch64_gen_load_pair (XEXP (pats[0], 0),
> -				      XEXP (pats[1], 0),
> -				      XEXP (pats[0], 1));
> -  else
> -    pair_pat = aarch64_gen_store_pair (XEXP (pats[0], 0),
> -				       XEXP (pats[0], 1),
> -				       XEXP (pats[1], 1));
> -
> +  rtx pair_pat = m_pass->gen_pair (pats, writeback_effect, load_p);
>    insn_change *pair_change = nullptr;
>    auto set_pair_pat = [pair_pat,reg_notes](insn_change *change) {
>      rtx_insn *rti = change->insn ()->rtl ();
> @@ -2125,15 +2341,6 @@ load_modified_by_store_p (insn_info *load,
>    return false;
>  }
>  
> -// Virtual base class for load/store walkers used in alias analysis.
> -struct alias_walker
> -{
> -  virtual bool conflict_p (int &budget) const = 0;
> -  virtual insn_info *insn () const = 0;
> -  virtual bool valid () const = 0;
> -  virtual void advance () = 0;
> -};
> -
>  // Implement some common functionality used by both store_walker
>  // and load_walker.
>  template<bool reverse>
> @@ -2251,13 +2458,13 @@ public:
>  //
>  // We try to maintain the invariant that if a walker becomes invalid, we
>  // set its pointer to null.
> -static void
> -do_alias_analysis (insn_info *alias_hazards[4],
> +void
> +pair_fusion::do_alias_analysis (insn_info *alias_hazards[4],
>  		   alias_walker *walkers[4],
>  		   bool load_p)
>  {
>    const int n_walkers = 2 + (2 * !load_p);
> -  int budget = aarch64_ldp_alias_check_limit;
> +  int budget = pair_mem_alias_check_limit ();
>  
>    auto next_walker = [walkers,n_walkers](int current) -> int {
>      for (int j = 1; j <= n_walkers; j++)
> @@ -2342,8 +2549,8 @@ do_alias_analysis (insn_info *alias_hazards[4],
>  //
>  // Returns an integer where bit (1 << i) is set if INSNS[i] uses writeback
>  // addressing.
> -static int
> -get_viable_bases (insn_info *insns[2],
> +int
> +pair_fusion::get_viable_bases (insn_info *insns[2],
>  		  vec<base_cand> &base_cands,
>  		  rtx cand_mems[2],
>  		  unsigned access_size,

Indentation looks off here.

> @@ -2389,7 +2596,7 @@ get_viable_bases (insn_info *insns[2],
>        if (!is_lower)
>  	base_off--;
>  
> -      if (base_off < LDP_MIN_IMM || base_off > LDP_MAX_IMM)
> +      if (!pair_mem_in_range_p (base_off))
>  	continue;
>  
>        use_info *use = find_access (insns[i]->uses (), REGNO (base));
> @@ -2446,7 +2653,7 @@ get_viable_bases (insn_info *insns[2],
>  }
>  
>  // Given two adjacent memory accesses of the same size, I1 and I2, try
> -// and see if we can merge them into a ldp or stp.
> +// and see if we can merge them into a paired access.
>  //
>  // ACCESS_SIZE gives the (common) size of a single access, LOAD_P is true
>  // if the accesses are both loads, otherwise they are both stores.
> @@ -2486,7 +2693,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>      {
>        if (dump_file)
>  	fprintf (dump_file,
> -		 "punting on ldp due to reg conflcits (%d,%d)\n",
> +		 "punting on pair mem load  due to reg conflcits (%d,%d)\n",

Nit: double space after "load".  I suppose just "load pair" would do here.

>  		 insns[0]->uid (), insns[1]->uid ());
>        return false;
>      }
> @@ -2504,7 +2711,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>  
>    auto_vec<base_cand, 2> base_cands (2);
>  
> -  int writeback = get_viable_bases (insns, base_cands, cand_mems,
> +  int writeback = m_pass->get_viable_bases (insns, base_cands, cand_mems,
>  				    access_size, reversed);

Indentation is wrong here.

>    if (base_cands.is_empty ())
>      {
> @@ -2633,7 +2840,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>      walkers[1] = &backward_store_walker;
>  
>    if (load_p && (mem_defs[0] || mem_defs[1]))
> -    do_alias_analysis (alias_hazards, walkers, load_p);
> +    m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
>    else
>      {
>        // We want to find any loads hanging off the first store.
> @@ -2642,7 +2849,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>        load_walker<true> backward_load_walker (mem_defs[1], insns[1], insns[0]);
>        walkers[2] = &forward_load_walker;
>        walkers[3] = &backward_load_walker;
> -      do_alias_analysis (alias_hazards, walkers, load_p);
> +      m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
>        // Now consolidate hazards back down.
>        if (alias_hazards[2]
>  	  && (!alias_hazards[0] || (*alias_hazards[2] < *alias_hazards[0])))
> @@ -2956,26 +3163,6 @@ ldp_bb_info::transform ()
>    traverse_base_map (def_map);
>  }
>  
> -static void
> -ldp_fusion_init ()
> -{
> -  calculate_dominance_info (CDI_DOMINATORS);
> -  df_analyze ();
> -  crtl->ssa = new rtl_ssa::function_info (cfun);
> -}
> -
> -static void
> -ldp_fusion_destroy ()
> -{
> -  if (crtl->ssa->perform_pending_updates ())
> -    cleanup_cfg (0);
> -
> -  free_dominance_info (CDI_DOMINATORS);
> -
> -  delete crtl->ssa;
> -  crtl->ssa = nullptr;
> -}
> -
>  // Given a load pair insn in PATTERN, unpack the insn, storing
>  // the registers in REGS and returning the mem.
>  static rtx
> @@ -3015,16 +3202,19 @@ aarch64_destructure_store_pair (rtx regs[2], rtx pattern)
>    return mem;
>  }
>  
> -// Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx
> -// representing the effect of writeback on the base register in WB_EFFECT,
> -// return an insn representing a writeback variant of this pair.
> -// LOAD_P is true iff the pair is a load.
> -//
> -// This is used when promoting existing non-writeback pairs to writeback
> -// variants.
> -static rtx
> -aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
> -			    bool load_p)
> +rtx
> +aarch64_pair_fusion::destructure_pair (rtx regs[2], rtx rti, bool load_p)
> +{
> +  if (load_p)
> +    return aarch64_destructure_load_pair (regs, rti);
> +  else
> +    return aarch64_destructure_store_pair (regs, rti);
> +}
> +
> +rtx
> +aarch64_pair_fusion::gen_promote_writeback_pair (rtx wb_effect, rtx pair_mem,
> +						 rtx regs[2],
> +						 bool load_p)
>  {
>    auto op_mode = aarch64_operand_mode_for_pair_mode (GET_MODE (pair_mem));
>  
> @@ -3059,23 +3249,13 @@ aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
>  // Given an existing pair insn INSN, look for a trailing update of
>  // the base register which we can fold in to make this pair use
>  // a writeback addressing mode.
> -static void
> -try_promote_writeback (insn_info *insn)
> +void
> +pair_fusion::try_promote_writeback (insn_info *insn, bool load_p)
>  {
> -  auto rti = insn->rtl ();
> -  const auto attr = get_attr_ldpstp (rti);
> -  if (attr == LDPSTP_NONE)
> -    return;
> -
> -  bool load_p = (attr == LDPSTP_LDP);
> -  gcc_checking_assert (load_p || attr == LDPSTP_STP);
> -
>    rtx regs[2];
>    rtx mem = NULL_RTX;
> -  if (load_p)
> -    mem = aarch64_destructure_load_pair (regs, PATTERN (rti));
> -  else
> -    mem = aarch64_destructure_store_pair (regs, PATTERN (rti));
> +
> +  mem = destructure_pair (regs, PATTERN (insn->rtl ()), load_p);

Sorry for missing this before, but it's better just to have:

  rtx mem = destructure_pair (...);

i.e. declare and initialize it on the same line.

>    gcc_checking_assert (MEM_P (mem));
>  
>    poly_int64 offset;
> @@ -3112,9 +3292,10 @@ try_promote_writeback (insn_info *insn)
>    def_info *add_def;
>    const insn_range_info pair_range (insn);
>    insn_info *insns[2] = { nullptr, insn };
> -  insn_info *trailing_add = find_trailing_add (insns, pair_range, 0, &wb_effect,
> -					       &add_def, base_def, offset,
> -					       access_size);
> +  insn_info *trailing_add
> +    = find_trailing_add (insns, pair_range, 0, &wb_effect,
> +			 &add_def, base_def, offset,
> +			 access_size);
>    if (!trailing_add)
>      return;
>  
> @@ -3124,8 +3305,10 @@ try_promote_writeback (insn_info *insn)
>    insn_change del_change (trailing_add, insn_change::DELETE);
>    insn_change *changes[] = { &pair_change, &del_change };
>  
> -  rtx pair_pat = aarch64_gen_writeback_pair (wb_effect, mem, regs, load_p);
> -  validate_unshare_change (rti, &PATTERN (rti), pair_pat, true);
> +  rtx pair_pat = gen_promote_writeback_pair (wb_effect, mem, regs,
> +					     load_p);

The params should all fit on a single line here.

> +  validate_unshare_change (insn->rtl(), &PATTERN (insn->rtl()), pair_pat,

Space before '(' in function calls (two occurrences here).

> +			   true);
>  
>    // The pair must gain the def of the base register from the add.
>    pair_change.new_defs = insert_access (attempt,
> @@ -3159,14 +3342,12 @@ try_promote_writeback (insn_info *insn)
>  // for load/store candidates.  If running after RA, also try and promote
>  // non-writeback pairs to use writeback addressing.  Then try to fuse
>  // candidates into pairs.
> -void ldp_fusion_bb (bb_info *bb)
> +void pair_fusion::ldp_fusion_bb (bb_info *bb)
>  {
> -  const bool track_loads
> -    = aarch64_tune_params.ldp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
> -  const bool track_stores
> -    = aarch64_tune_params.stp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
> +  const bool track_loads = track_loads_p ();
> +  const bool track_stores = track_stores_p ();
>  
> -  ldp_bb_info bb_state (bb);
> +  ldp_bb_info bb_state (bb, this);
>  
>    for (auto insn : bb->nondebug_insns ())
>      {
> @@ -3176,11 +3357,11 @@ void ldp_fusion_bb (bb_info *bb)
>  	continue;
>  
>        rtx pat = PATTERN (rti);
> +      bool load_p;
>        if (reload_completed
> -	  && aarch64_ldp_writeback > 1
> -	  && GET_CODE (pat) == PARALLEL
> -	  && XVECLEN (pat, 0) == 2)
> -	try_promote_writeback (insn);
> +	  && handle_writeback_opportunities (writeback::ALL)
> +	  && pair_mem_insn_p (rti, load_p))
> +	try_promote_writeback (insn, load_p);
>  
>        if (GET_CODE (pat) != SET)
>  	continue;
> @@ -3195,16 +3376,6 @@ void ldp_fusion_bb (bb_info *bb)
>    bb_state.cleanup_tombstones ();
>  }
>  
> -void ldp_fusion ()
> -{
> -  ldp_fusion_init ();
> -
> -  for (auto bb : crtl->ssa->bbs ())
> -    ldp_fusion_bb (bb);
> -
> -  ldp_fusion_destroy ();
> -}
> -
>  namespace {
>  
>  const pass_data pass_data_ldp_fusion =
> @@ -3234,14 +3405,6 @@ public:
>        if (!optimize || optimize_debug)
>  	return false;
>  
> -      // If the tuning policy says never to form ldps or stps, don't run
> -      // the pass.
> -      if ((aarch64_tune_params.ldp_policy_model
> -	   == AARCH64_LDP_STP_POLICY_NEVER)
> -	  && (aarch64_tune_params.stp_policy_model
> -	      == AARCH64_LDP_STP_POLICY_NEVER))
> -	return false;
> -
>        if (reload_completed)
>  	return flag_aarch64_late_ldp_fusion;
>        else
> @@ -3250,7 +3413,8 @@ public:
>  
>    unsigned execute (function *) final override
>      {
> -      ldp_fusion ();
> +      aarch64_pair_fusion pass;
> +      pass.run ();
>        return 0;
>      }
>  };
> -- 
> 2.39.3
> 

Thanks,
Alex
Ajit Agarwal May 15, 2024, 9:48 a.m. UTC | #2
Hello Alex:

On 14/05/24 11:53 pm, Alex Coplan wrote:
> Hi Ajit,
> 
> Please can you pay careful attention to the review comments?
> 
> In particular, you have ignored my comment about changing the access of
> member functions in ldp_bb_info several times now (on at least three
> patch reviews).
> 
> Likewise on multiple occasions you've only partially implemented a piece
> of review feedback (e.g. applying the "override" keyword to virtual
> overrides).
> 
> That all makes it rather tiresome to review your patches.
> 
> Also, I realise I should have mentioned this on a previous revision of
> this patch, but I thought we previously agreed (with Richard S) to split
> out the renaming in existing code (e.g. ldp/stp -> "paired access" and
> so on) to a separate patch?  That would make this eaiser to review.
> 

Sorry for the inconvenience caused. Hopefully I have incorporated
all the comments in v6 version of the patch.

> On 14/05/2024 15:08, Ajit Agarwal wrote:
>> Hello Alex/Richard:
>>
>> All comments are addressed.
>>
>> Common infrastructure of load store pair fusion is divided into target
>> independent and target dependent changed code.
>>
>> Target independent code is the Generic code with pure virtual function
>> to interface betwwen target independent and dependent code.
>>
>> Target dependent code is the implementation of pure virtual function for
>> aarch64 target and the call to target independent code.
>>
>> Bootstrapped on aarch64-linux-gnu.
>>
>> Thanks & Regards
>> Ajit
>>
>>
>>
>> arch64: Preparatory patch to place target independent and
>> dependent changed code in one file
>>
>> Common infrastructure of load store pair fusion is divided into target
>> independent and target dependent changed code.
>>
>> Target independent code is the Generic code with pure virtual function
>> to interface betwwen target independent and dependent code.
>>
>> Target dependent code is the implementation of pure virtual function for
>> aarch64 target and the call to target independent code.
>>
>> 2024-05-14  Ajit Kumar Agarwal  <aagarwa1@linux.ibm.com>
>>
>> gcc/ChangeLog:
>>
>> 	* config/aarch64/aarch64-ldp-fusion.cc: Place target
>> 	independent and dependent changed code.
>> ---
>>  gcc/config/aarch64/aarch64-ldp-fusion.cc | 526 +++++++++++++++--------
>>  1 file changed, 345 insertions(+), 181 deletions(-)
>>
>> diff --git a/gcc/config/aarch64/aarch64-ldp-fusion.cc b/gcc/config/aarch64/aarch64-ldp-fusion.cc
>> index 1d9caeab05d..e6af4b0570a 100644
>> --- a/gcc/config/aarch64/aarch64-ldp-fusion.cc
>> +++ b/gcc/config/aarch64/aarch64-ldp-fusion.cc
>> @@ -138,6 +138,210 @@ struct alt_base
>>    poly_int64 offset;
>>  };
>>  
>> +// Virtual base class for load/store walkers used in alias analysis.
>> +struct alias_walker
>> +{
>> +  virtual bool conflict_p (int &budget) const = 0;
>> +  virtual insn_info *insn () const = 0;
>> +  virtual bool valid () const = 0;
>> +  virtual void advance () = 0;
>> +};
>> +
>> +// This is used in handle_writeback_opportunities describing
>> +// ALL if aarch64_ldp_writeback > 1 otherwise check
>> +// EXISTING if aarch64_ldp_writeback.
> 
> Since this enum belongs to the generic interface, it's best if it is
> described in general terms, i.e. the comment shouldn't refer to the
> aarch64 param.
> 
> How about:
> 
> // When querying handle_writeback_opportunities, this enum is used to
> // qualify which opportunities we are asking about.
> 
> then above the EXISTING enumerator, you could say:
> 
>   // Only those writeback opportunities that arise from existing
>   // auto-increment accesses.
> 
> and for ALL, you could say:
> 
>   // All writeback opportunities including those that involve folding
>   // base register updates into a non-writeback pair.
>

Addressed in v6 of the patch.
 
>> +enum class writeback {
>> +  ALL,
>> +  EXISTING
>> +};
> 
> Also, sorry for the very minor nit, but I think it is more logical if we
> flip the order of the enumerators here, i.e. EXISTING should come first.
> 
>> +
>> +struct pair_fusion {
>> +  pair_fusion ()
>> +  {
>> +    calculate_dominance_info (CDI_DOMINATORS);
>> +    df_analyze ();
>> +    crtl->ssa = new rtl_ssa::function_info (cfun);
>> +  };
>> +
>> +  // Given:
>> +  // - an rtx REG_OP, the non-memory operand in a load/store insn,
>> +  // - a machine_mode MEM_MODE, the mode of the MEM in that insn, and
>> +  // - a boolean LOAD_P (true iff the insn is a load), then:
>> +  // return true if the access should be considered an FP/SIMD access.
>> +  // Such accesses are segregated from GPR accesses, since we only want
>> +  // to form pairs for accesses that use the same register file.
>> +  virtual bool fpsimd_op_p (rtx, machine_mode, bool)
>> +  {
>> +    return false;
>> +  }
>> +
>> +  // Return true if we should consider forming ldp/stp insns from memory
> 
> Replace "ldp/stp insns" with "pairs" here, since this is the generic
> interface.
> 

Addressed in v6 of the patch.
>> +  // accesses with operand mode MODE at this stage in compilation.
>> +  virtual bool pair_operand_mode_ok_p (machine_mode mode) = 0;
>> +
>> +  // Return true iff REG_OP is a suitable register operand for a paired
>> +  // memory access, where LOAD_P is true if we're asking about loads and
>> +  // false for stores.  MODE gives the mode of the operand.
>> +  virtual bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
>> +				      machine_mode mode) = 0;
>> +
>> +  // Return alias check limit.
>> +  // This is needed to avoid unbounded quadratic behaviour when
>> +  // performing alias analysis.
>> +  virtual int pair_mem_alias_check_limit () = 0;
>> +
>> +  // Returns true if we should try to handle writeback opportunities
> 
> Missing punctuation at the end of the sentence here.
>

Addressed in v6 of the patch.
 
>> +  // WHICH parameter decides ALL or EXISTING writeback pairs.
> 
> How about: "WHICH determines the kinds of writeback opportunities the
> caller is asking about."
> 

Addressed in v6 of the patch.
>> +  virtual bool handle_writeback_opportunities (enum writeback which) = 0 ;
>> +
>> +  // Given BASE_MEM, the mem from the lower candidate access for a pair,
>> +  // and LOAD_P (true if the access is a load), check if we should proceed
>> +  // to form the pair given the target's code generation policy on
>> +  // paired accesses.
>> +  virtual bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) = 0;
>> +
>> +  // Generate the pattern for a paired access. PATS gives the patterns
> 

Addressed in v6 of the patch.
> Nit: two spaces after '.'.  Run your patch through
> contrib/check_GNU_style.py before submitting, please.
>

Addressed in v6 of the patch.
 
>> +  // for the individual memory accesses (which by this point must share a
>> +  // common base register).  If WRITEBACK is non-NULL, then this rtx
>> +  // describes the update to the base register that should be performed by
>> +  // the resulting insn.  LOAD_P is true iff the accesses are loads.
>> +  virtual rtx gen_pair (rtx *pats, rtx writeback, bool load_p) = 0;
>> +
>> +  // Return true if memory is paired access, given INSN and LOAD_P
>> +  // is true for load insn and false for store insn.
>> +  virtual bool pair_mem_insn_p (rtx_insn *, bool &) = 0;
> 
> Since this is now a pure virtual, you can use parameter names in the
> prototype, then refer to them directly in the comment, so something
> like:
> 

Addressed in v6 of the patch.
> // Return true if INSN is a paired memory access.  If so, set LOAD_P to
> // true iff INSN is a load pair.
> 

Addressed in v6 of the patch.
>> +
>> +  // Return true if we should track loads.
>> +  virtual bool track_loads_p ()
>> +  {
>> +    return true;
>> +  }
>> +
>> +  // Return true if we should track stores.
>> +  virtual bool track_stores_p ()
>> +  {
>> +    return true;
>> +  }
>> +
>> +  // Return true if OFF  is in range.
> 
Addressed in v6 of the patch.
> Excess space after OFF.  Sorry for missing this last time, but "in
> range" by itself isn't meaningful, how about "in range for a paired
> memory access" instead.
> 
Addressed in v6 of the patch.
>> +  virtual bool pair_mem_in_range_p (HOST_WIDE_INT off) = 0;
> 
> Sorry for the nit, but I suggested last time renaming s/off/offset/.
> Let's do that, please.
> 
Addressed in v6 of the patch.
>> +
>> +  // Given a load/store pair insn in PATTERN, unpack the insn, storing
>> +  // the register operands in REGS, and returning the mem. LOAD_P is true
> 
> Nit: two spaces after '.'.
> 

Addressed in v6 of the patch.
>> +  // for load and false for store.
> 
> Sorry for the nit, but I think the last sentence should say "LOAD_P is
> true for loads and false for stores." (i.e. plural loads/stores).
> 
Addressed in v6 of the patch.

>> +  virtual rtx destructure_pair (rtx regs[2], rtx pattern, bool load_p) = 0;
>> +
>> +  // Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx
> 
> The name PAIR_MEM doesn't match the parameter name.  One of them needs
> to change.
> 

Addressed in v6 of the patch.
>> +  // representing the effect of writeback on the base register in WB_EFFECT,
>> +  // return an insn representing a writeback variant of this pair.
>> +  // LOAD_P is true iff the pair is a load.
>> +  // This is used when promoting existing non-writeback pairs to writeback
>> +  // variants.
>> +  virtual rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem,
>> +					  rtx regs[2], bool load_p) = 0;
>> +
>> +  void ldp_fusion_bb (bb_info *bb);
>> +  inline insn_info *find_trailing_add (insn_info *insns[2],
>> +				       const insn_range_info &pair_range,
>> +				       int initial_writeback,
>> +				       rtx *writeback_effect,
>> +				       def_info **add_def,
>> +				       def_info *base_def,
>> +				       poly_int64 initial_offset,
>> +				       unsigned access_size);
>> +  inline int get_viable_bases (insn_info *insns[2],
>> +			       vec<base_cand> &base_cands,
>> +			       rtx cand_mems[2],
>> +			       unsigned access_size,
>> +			       bool reversed);
>> +  inline void do_alias_analysis (insn_info *alias_hazards[4],
>> +				 alias_walker *walkers[4],
>> +				 bool load_p);
>> +  inline void try_promote_writeback (insn_info *insn, bool load_p);
>> +  inline void run ();
>> +  ~pair_fusion()
> 
Addressed in v6 of the patch.
> Nit: space before '('.  Again, contrib/check_GNU_style.py can help here.
> 
>> +  {
>> +    if (crtl->ssa->perform_pending_updates ())
>> +      cleanup_cfg (0);
>> +
>> +    free_dominance_info (CDI_DOMINATORS);
>> +
>> +    delete crtl->ssa;
>> +    crtl->ssa = nullptr;
>> +  }
>> +};
>> +
>> +// This is the main function to start the pass.
>> +void
>> +pair_fusion::run ()
>> +{
>> +  if (!track_loads_p () && !track_stores_p ())
>> +    return;
>> +
>> +  for (auto bb : crtl->ssa->bbs ())
>> +    ldp_fusion_bb (bb);
>> +}
>> +
>> +struct aarch64_pair_fusion : public pair_fusion
>> +{
>> +  // Before RA, we use the modes, noting that stores of constant zero
>> +  // operands use GPRs (even in non-integer modes).  After RA, we use
>> +  // the hard register numbers.
> 

Addressed in v6 of the patch.
> Sorry for the very minor nit, but I think this comment would fit better
> inside the function body (above the return statement).
> 

Addressed in v6 of the patch.
>> +  bool fpsimd_op_p (rtx reg_op, machine_mode mem_mode,
>> +		    bool load_p) override final
>> +  {
>> +    return reload_completed
>> +      ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
>> +      : (GET_MODE_CLASS (mem_mode) != MODE_INT
>> +	 && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
>> +  }
>> +
>> +  bool pair_mem_insn_p (rtx_insn *rti, bool &load_p) override final;
>> +  bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) override final
>> +  {
>> +    return aarch64_mem_ok_with_ldpstp_policy_model (base_mem,
>> +						    load_p,
>> +						    GET_MODE (base_mem));
>> +  }
> 
Addressed in v6 of the patch.

> Can we add blank lines between all of the inline member function
> implementations in this class, please?  I did ask about that last time.
> 
>> +  bool pair_operand_mode_ok_p (machine_mode mode);
> 
> Looks like you're still missing "override final" on all of these
> remaining overrides.
> 
>> +  rtx gen_pair (rtx *pats, rtx writeback, bool load_p);
>> +  bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
>> +			      machine_mode mode)
>> +  {
>> +    return (load_p
>> +	    ? aarch64_ldp_reg_operand (reg_op, mode)
>> +	    : aarch64_stp_reg_operand (reg_op, mode));
>> +  }
>> +  int pair_mem_alias_check_limit ()
>> +  {
>> +    return aarch64_ldp_alias_check_limit;
>> +  }
>> +  bool handle_writeback_opportunities (enum writeback which)
>> +  {
>> +    if (which == writeback::ALL)
>> +      return aarch64_ldp_writeback > 1;
>> +    else
>> +      return aarch64_ldp_writeback;
>> +  }
>> +  bool track_loads_p ()
>> +  {
>> +    return aarch64_tune_params.ldp_policy_model
>> +	   != AARCH64_LDP_STP_POLICY_NEVER;
>> +  }
>> +  bool track_stores_p ()
>> +  {
>> +    return aarch64_tune_params.stp_policy_model
>> +	   != AARCH64_LDP_STP_POLICY_NEVER;
>> +  }
>> +  bool pair_mem_in_range_p (HOST_WIDE_INT off)
>> +  {
>> +    return (off >= LDP_MIN_IMM && off <= LDP_MAX_IMM);
>> +  }
>> +  rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem, rtx regs[2],
>> +				  bool load_p);
>> +  rtx destructure_pair (rtx regs[2], rtx rti, bool load_p);
>> +};
>> +
>>  // State used by the pass for a given basic block.
>>  struct ldp_bb_info
>>  {
>> @@ -159,9 +363,9 @@ struct ldp_bb_info
>>    hash_map<def_hash, alt_base> canon_base_map;
>>  
>>    static const size_t obstack_alignment = sizeof (void *);
>> -  bb_info *m_bb;
>>  
>> -  ldp_bb_info (bb_info *bb) : m_bb (bb), m_emitted_tombstone (false)
>> +  ldp_bb_info (bb_info *bb, pair_fusion *d)
>> +	       : m_bb (bb), m_pass (d), m_emitted_tombstone (false)
>>    {
>>      obstack_specify_allocation (&m_obstack, OBSTACK_CHUNK_SIZE,
>>  				obstack_alignment, obstack_chunk_alloc,
>> @@ -181,37 +385,75 @@ struct ldp_bb_info
>>    inline void track_access (insn_info *, bool load, rtx mem);
>>    inline void transform ();
>>    inline void cleanup_tombstones ();
>> +  inline void merge_pairs (insn_list_t &, insn_list_t &,
>> +		    bool load_p, unsigned access_size);
>> +  inline void transform_for_base (int load_size, access_group &group);
>> +
>> +  inline bool try_fuse_pair (bool load_p, unsigned access_size,
>> +			     insn_info *i1, insn_info *i2);
>> +
>> +  inline bool fuse_pair (bool load_p, unsigned access_size,
>> +		  int writeback,
>> +		  insn_info *i1, insn_info *i2,
>> +		  base_cand &base,
>> +		  const insn_range_info &move_range);
>> +
>> +  inline void track_tombstone (int uid);
> 
> You keep missing my review comments here.  Why are you changing these
> member functions from private to public?
> 

Addressed in v6 of the patch.
>>  
>>  private:
>>    obstack m_obstack;
>> -
> 
> Can we keep the blank lines here?  They were there to logically group
> together the members related to tracking tombstones.
> 
>> +  bb_info *m_bb;
>> +  pair_fusion *m_pass;
>>    // State for keeping track of tombstone insns emitted for this BB.
>>    bitmap_obstack m_bitmap_obstack;
>>    bitmap_head m_tombstone_bitmap;
>>    bool m_emitted_tombstone;
>> -
>> +  template<typename Map> inline void traverse_base_map (Map &map);
>> +  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
>>    inline splay_tree_node<access_record *> *node_alloc (access_record *);
>> +};
> 
> What is the purpose of these changes?  Why are you moving these
> function declarations around?
> 

Addressed in v6 of the patch.
>>  
>> -  template<typename Map>
>> -  inline void traverse_base_map (Map &map);
>> -  inline void transform_for_base (int load_size, access_group &group);
>> -
>> -  inline void merge_pairs (insn_list_t &, insn_list_t &,
>> -			   bool load_p, unsigned access_size);
>> -
>> -  inline bool try_fuse_pair (bool load_p, unsigned access_size,
>> -			     insn_info *i1, insn_info *i2);
>> +bool
>> +aarch64_pair_fusion::pair_mem_insn_p (rtx_insn *rti,
>> +				      bool &load_p)
> 
> No need for the line break here, the params should fit on a single line.
> 
>> +{
>> +  rtx pat = PATTERN (rti);
>> +  if (GET_CODE (pat) == PARALLEL
>> +      && XVECLEN (pat, 0) == 2)
>> +    {
>> +      const auto attr = get_attr_ldpstp (rti);
>> +      if (attr == LDPSTP_NONE)
>> +	return false;
>>  
>> -  inline bool fuse_pair (bool load_p, unsigned access_size,
>> -			 int writeback,
>> -			 insn_info *i1, insn_info *i2,
>> -			 base_cand &base,
>> -			 const insn_range_info &move_range);
>> +      load_p = (attr == LDPSTP_LDP);
>> +      gcc_checking_assert (load_p || attr == LDPSTP_STP);
>> +      return true;
>> +    }
>> +  return false;
>> +}
>>  
>> -  inline void track_tombstone (int uid);
>> +rtx
>> +aarch64_pair_fusion::gen_pair (rtx *pats,
>> +			       rtx writeback,
>> +			       bool load_p)
> 
Addressed in v6 of the patch.

> Likewise, the params should all fit on a single line.
> 
>> +{
>> +  rtx pair_pat;
>>  
>> -  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
>> -};
>> +  if (writeback)
>> +    {
>> +      auto patvec = gen_rtvec (3, writeback, pats[0], pats[1]);
>> +      return  gen_rtx_PARALLEL (VOIDmode, patvec);
> 
> Excess space after return.
> 
Addressed in v6 of the patch.

>> +    }
>> +  else if (load_p)
>> +    return aarch64_gen_load_pair (XEXP (pats[0], 0),
>> +				  XEXP (pats[1], 0),
>> +				  XEXP (pats[0], 1));
>> +  else
>> +    return aarch64_gen_store_pair (XEXP (pats[0], 0),
>> +				   XEXP (pats[0], 1),
>> +				   XEXP (pats[1], 1));
>> +  return pair_pat;
>> +}
>>  
>>  splay_tree_node<access_record *> *
>>  ldp_bb_info::node_alloc (access_record *access)
>> @@ -312,8 +554,8 @@ any_post_modify_p (rtx x)
>>  
>>  // Return true if we should consider forming ldp/stp insns from memory
>>  // accesses with operand mode MODE at this stage in compilation.
>> -static bool
>> -ldp_operand_mode_ok_p (machine_mode mode)
>> +bool
>> +aarch64_pair_fusion::pair_operand_mode_ok_p (machine_mode mode)
>>  {
>>    if (!aarch64_ldpstp_operand_mode_p (mode))
>>      return false;
>> @@ -404,9 +646,10 @@ ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
>>    const machine_mode mem_mode = GET_MODE (mem);
>>    const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
>>  
>> -  // Punt on misaligned offsets.  LDP/STP instructions require offsets to be a
>> -  // multiple of the access size, and we believe that misaligned offsets on
>> -  // MEM_EXPR bases are likely to lead to misaligned offsets w.r.t. RTL bases.
>> +  // Punt on misaligned offsets.  Paired memory access instructions require
>> +  // offsets to be a multiple of the access size, and we believe that
>> +  // misaligned offsets on MEM_EXPR bases are likely to lead to misaligned
>> +  // offsets w.r.t. RTL bases.
>>    if (!multiple_p (offset, mem_size))
>>      return false;
>>  
>> @@ -430,10 +673,10 @@ ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
>>  }
>>  
>>  // Main function to begin pair discovery.  Given a memory access INSN,
>> -// determine whether it could be a candidate for fusing into an ldp/stp,
>> -// and if so, track it in the appropriate data structure for this basic
>> -// block.  LOAD_P is true if the access is a load, and MEM is the mem
>> -// rtx that occurs in INSN.
>> +// determine whether it could be a candidate for fusing into a paired
>> +// access, and if so, track it in the appropriate data structure for
>> +// this basic block.  LOAD_P is true if the access is a load, and MEM
>> +// is the mem rtx that occurs in INSN.
>>  void
>>  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>>  {
>> @@ -441,35 +684,24 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>>    if (MEM_VOLATILE_P (mem))
>>      return;
>>  
>> -  // Ignore writeback accesses if the param says to do so.
>> -  if (!aarch64_ldp_writeback
>> +  // Ignore writeback accesses if the hook says to do so.
>> +  if (!m_pass->handle_writeback_opportunities (writeback::EXISTING)
>>        && GET_RTX_CLASS (GET_CODE (XEXP (mem, 0))) == RTX_AUTOINC)
>>      return;
>>  
>>    const machine_mode mem_mode = GET_MODE (mem);
>> -  if (!ldp_operand_mode_ok_p (mem_mode))
>> +  if (!m_pass->pair_operand_mode_ok_p (mem_mode))
>>      return;
>>  
>>    rtx reg_op = XEXP (PATTERN (insn->rtl ()), !load_p);
>>  
>> -  // Ignore the access if the register operand isn't suitable for ldp/stp.
>> -  if (load_p
>> -      ? !aarch64_ldp_reg_operand (reg_op, mem_mode)
>> -      : !aarch64_stp_reg_operand (reg_op, mem_mode))
>> +  if (!m_pass->pair_reg_operand_ok_p (load_p, reg_op, mem_mode))
>>      return;
>>  
>>    // We want to segregate FP/SIMD accesses from GPR accesses.
>> -  //
>> -  // Before RA, we use the modes, noting that stores of constant zero
>> -  // operands use GPRs (even in non-integer modes).  After RA, we use
>> -  // the hard register numbers.
>> -  const bool fpsimd_op_p
>> -    = reload_completed
>> -    ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
>> -    : (GET_MODE_CLASS (mem_mode) != MODE_INT
>> -       && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
>> +  const bool fpsimd_op_p = m_pass->fpsimd_op_p (reg_op, mem_mode, load_p);
>>  
>> -  // Note ldp_operand_mode_ok_p already rejected VL modes.
>> +  // Note pair_operand_mode_ok_p already rejected VL modes.
>>    const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
>>    const lfs_fields lfs = { load_p, fpsimd_op_p, mem_size };
>>  
>> @@ -498,8 +730,8 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>>    // elimination offset pre-RA, we should postpone forming pairs on such
>>    // accesses until after RA.
>>    //
>> -  // As it stands, addresses with offsets in range for LDR but not
>> -  // in range for LDP/STP are currently reloaded inefficiently,
>> +  // As it stands, addresses in range for an individual load/store but not
>> +  // for a paired access are currently reloaded inefficiently,
>>    // ending up with a separate base register for each pair.
>>    //
>>    // In theory LRA should make use of
>> @@ -510,11 +742,11 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>>    // constraint; this early return means we never get to the code
>>    // that calls targetm.legitimize_address_displacement.
>>    //
>> -  // So for now, it's better to punt when we can't be sure that the
>> -  // offset is in range for LDP/STP.  Out-of-range cases can then be
>> -  // handled after RA by the out-of-range LDP/STP peepholes.  Eventually, it
>> -  // would be nice to handle known out-of-range opportunities in the
>> -  // pass itself (for stack accesses, this would be in the post-RA pass).
>> +  // It's better to punt when we can't be sure that the
>> +  // offset is in range for paired access.  On aarch64, Out-of-range cases
>> +  // can then be handled after RA by the out-of-range PAIR MEM  peepholes.
> 
> No need to change "LDP/STP" to "PAIR MEM " here, now that you've
> qualified that sentence as being aarch64-specific.
> 
Addressed in v6 of the patch.

>> +  // Eventually, it would be nice to handle known out-of-range opportunities
>> +  // in the pass itself (for stack accesses, this would be in the post-RA pass).
>>    if (!reload_completed
>>        && (REGNO (base) == FRAME_POINTER_REGNUM
>>  	  || REGNO (base) == ARG_POINTER_REGNUM))
>> @@ -565,8 +797,8 @@ ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
>>  	gcc_unreachable (); // Base defs should be unique.
>>      }
>>  
>> -  // Punt on misaligned offsets.  LDP/STP require offsets to be a multiple of
>> -  // the access size.
>> +  // Punt on misaligned offsets.  Paired memory accesses require offsets
>> +  // to be a multiple of the access size.
>>    if (!multiple_p (mem_off, mem_size))
>>      return;
>>  
>> @@ -1199,8 +1431,8 @@ extract_writebacks (bool load_p, rtx pats[2], int changed)
>>  // base register.  If there is one, we choose the first such update after
>>  // PAIR_DST that is still in the same BB as our pair.  We return the new def in
>>  // *ADD_DEF and the resulting writeback effect in *WRITEBACK_EFFECT.
>> -static insn_info *
>> -find_trailing_add (insn_info *insns[2],
>> +insn_info *
>> +pair_fusion::find_trailing_add (insn_info *insns[2],
>>  		   const insn_range_info &pair_range,
>>  		   int initial_writeback,
>>  		   rtx *writeback_effect,
>> @@ -1278,7 +1510,7 @@ find_trailing_add (insn_info *insns[2],
>>  
>>    off_hwi /= access_size;
>>  
>> -  if (off_hwi < LDP_MIN_IMM || off_hwi > LDP_MAX_IMM)
>> +  if (!pair_mem_in_range_p (off_hwi))
>>      return nullptr;
>>  
>>    auto dump_prefix = [&]()
>> @@ -1792,7 +2024,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>>      {
>>        if (dump_file)
>>  	fprintf (dump_file,
>> -		 "  ldp: i%d has wb but subsequent i%d has non-wb "
>> +		 "  load pair: i%d has wb but subsequent i%d has non-wb "
>>  		 "update of base (r%d), dropping wb\n",
>>  		 insns[0]->uid (), insns[1]->uid (), base_regno);
>>        gcc_assert (writeback_effect);
>> @@ -1815,7 +2047,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>>      }
>>  
>>    // If either of the original insns had writeback, but the resulting pair insn
>> -  // does not (can happen e.g. in the ldp edge case above, or if the writeback
>> +  // does not (can happen e.g. in the load pair edge case above, or if the writeback
> 
> Nit: long line here (comment needs re-wrapping).
> 
Addressed in v6 of the patch.

>>    // effects cancel out), then drop the def(s) of the base register as
>>    // appropriate.
>>    //
>> @@ -1834,7 +2066,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>>    // update of the base register and try and fold it in to make this into a
>>    // writeback pair.
>>    insn_info *trailing_add = nullptr;
>> -  if (aarch64_ldp_writeback > 1
>> +  if (m_pass->handle_writeback_opportunities (writeback::ALL)
>>        && !writeback_effect
>>        && (!load_p || (!refers_to_regno_p (base_regno, base_regno + 1,
>>  					 XEXP (pats[0], 0), nullptr)
>> @@ -1842,7 +2074,7 @@ ldp_bb_info::fuse_pair (bool load_p,
>>  					     XEXP (pats[1], 0), nullptr))))
>>      {
>>        def_info *add_def;
>> -      trailing_add = find_trailing_add (insns, move_range, writeback,
>> +      trailing_add = m_pass->find_trailing_add (insns, move_range, writeback,
>>  					&writeback_effect,
>>  					&add_def, base.def, offsets[0],
>>  					access_size);
> 
> Looks like the parameters need re-indenting here.
> 
>> @@ -1855,35 +2087,19 @@ ldp_bb_info::fuse_pair (bool load_p,
>>      }
>>  
>>    // Now that we know what base mem we're going to use, check if it's OK
>> -  // with the ldp/stp policy.
>> +  // with the pair mem policy.
>>    rtx first_mem = XEXP (pats[0], load_p);
>> -  if (!aarch64_mem_ok_with_ldpstp_policy_model (first_mem,
>> -						load_p,
>> -						GET_MODE (first_mem)))
>> +  if (!m_pass->pair_mem_ok_with_policy (first_mem, load_p))
>>      {
>>        if (dump_file)
>> -	fprintf (dump_file, "punting on pair (%d,%d), ldp/stp policy says no\n",
>> +	fprintf (dump_file, "punting on pair (%d,%d), pair mem policy says no\n",
>>  		 i1->uid (), i2->uid ());
>>        return false;
>>      }
>>  
>>    rtx reg_notes = combine_reg_notes (first, second, load_p);
>>  
>> -  rtx pair_pat;
>> -  if (writeback_effect)
>> -    {
>> -      auto patvec = gen_rtvec (3, writeback_effect, pats[0], pats[1]);
>> -      pair_pat = gen_rtx_PARALLEL (VOIDmode, patvec);
>> -    }
>> -  else if (load_p)
>> -    pair_pat = aarch64_gen_load_pair (XEXP (pats[0], 0),
>> -				      XEXP (pats[1], 0),
>> -				      XEXP (pats[0], 1));
>> -  else
>> -    pair_pat = aarch64_gen_store_pair (XEXP (pats[0], 0),
>> -				       XEXP (pats[0], 1),
>> -				       XEXP (pats[1], 1));
>> -
>> +  rtx pair_pat = m_pass->gen_pair (pats, writeback_effect, load_p);
>>    insn_change *pair_change = nullptr;
>>    auto set_pair_pat = [pair_pat,reg_notes](insn_change *change) {
>>      rtx_insn *rti = change->insn ()->rtl ();
>> @@ -2125,15 +2341,6 @@ load_modified_by_store_p (insn_info *load,
>>    return false;
>>  }
>>  
>> -// Virtual base class for load/store walkers used in alias analysis.
>> -struct alias_walker
>> -{
>> -  virtual bool conflict_p (int &budget) const = 0;
>> -  virtual insn_info *insn () const = 0;
>> -  virtual bool valid () const = 0;
>> -  virtual void advance () = 0;
>> -};
>> -
>>  // Implement some common functionality used by both store_walker
>>  // and load_walker.
>>  template<bool reverse>
>> @@ -2251,13 +2458,13 @@ public:
>>  //
>>  // We try to maintain the invariant that if a walker becomes invalid, we
>>  // set its pointer to null.
>> -static void
>> -do_alias_analysis (insn_info *alias_hazards[4],
>> +void
>> +pair_fusion::do_alias_analysis (insn_info *alias_hazards[4],
>>  		   alias_walker *walkers[4],
>>  		   bool load_p)
>>  {
>>    const int n_walkers = 2 + (2 * !load_p);
>> -  int budget = aarch64_ldp_alias_check_limit;
>> +  int budget = pair_mem_alias_check_limit ();
>>  
>>    auto next_walker = [walkers,n_walkers](int current) -> int {
>>      for (int j = 1; j <= n_walkers; j++)
>> @@ -2342,8 +2549,8 @@ do_alias_analysis (insn_info *alias_hazards[4],
>>  //
>>  // Returns an integer where bit (1 << i) is set if INSNS[i] uses writeback
>>  // addressing.
>> -static int
>> -get_viable_bases (insn_info *insns[2],
>> +int
>> +pair_fusion::get_viable_bases (insn_info *insns[2],
>>  		  vec<base_cand> &base_cands,
>>  		  rtx cand_mems[2],
>>  		  unsigned access_size,
> 

Addressed in v6 of the patch.
> Indentation looks off here.
> 
>> @@ -2389,7 +2596,7 @@ get_viable_bases (insn_info *insns[2],
>>        if (!is_lower)
>>  	base_off--;
>>  
>> -      if (base_off < LDP_MIN_IMM || base_off > LDP_MAX_IMM)
>> +      if (!pair_mem_in_range_p (base_off))
>>  	continue;
>>  
>>        use_info *use = find_access (insns[i]->uses (), REGNO (base));
>> @@ -2446,7 +2653,7 @@ get_viable_bases (insn_info *insns[2],
>>  }
>>  
>>  // Given two adjacent memory accesses of the same size, I1 and I2, try
>> -// and see if we can merge them into a ldp or stp.
>> +// and see if we can merge them into a paired access.
>>  //
>>  // ACCESS_SIZE gives the (common) size of a single access, LOAD_P is true
>>  // if the accesses are both loads, otherwise they are both stores.
>> @@ -2486,7 +2693,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>>      {
>>        if (dump_file)
>>  	fprintf (dump_file,
>> -		 "punting on ldp due to reg conflcits (%d,%d)\n",
>> +		 "punting on pair mem load  due to reg conflcits (%d,%d)\n",
> 
> Nit: double space after "load".  I suppose just "load pair" would do here.
> 
>>  		 insns[0]->uid (), insns[1]->uid ());
>>        return false;
>>      }
>> @@ -2504,7 +2711,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>>  
>>    auto_vec<base_cand, 2> base_cands (2);
>>  
>> -  int writeback = get_viable_bases (insns, base_cands, cand_mems,
>> +  int writeback = m_pass->get_viable_bases (insns, base_cands, cand_mems,
>>  				    access_size, reversed);
> 
> Indentation is wrong here.
> 
Addressed in v6 of the patch.

>>    if (base_cands.is_empty ())
>>      {
>> @@ -2633,7 +2840,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>>      walkers[1] = &backward_store_walker;
>>  
>>    if (load_p && (mem_defs[0] || mem_defs[1]))
>> -    do_alias_analysis (alias_hazards, walkers, load_p);
>> +    m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
>>    else
>>      {
>>        // We want to find any loads hanging off the first store.
>> @@ -2642,7 +2849,7 @@ ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
>>        load_walker<true> backward_load_walker (mem_defs[1], insns[1], insns[0]);
>>        walkers[2] = &forward_load_walker;
>>        walkers[3] = &backward_load_walker;
>> -      do_alias_analysis (alias_hazards, walkers, load_p);
>> +      m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
>>        // Now consolidate hazards back down.
>>        if (alias_hazards[2]
>>  	  && (!alias_hazards[0] || (*alias_hazards[2] < *alias_hazards[0])))
>> @@ -2956,26 +3163,6 @@ ldp_bb_info::transform ()
>>    traverse_base_map (def_map);
>>  }
>>  
>> -static void
>> -ldp_fusion_init ()
>> -{
>> -  calculate_dominance_info (CDI_DOMINATORS);
>> -  df_analyze ();
>> -  crtl->ssa = new rtl_ssa::function_info (cfun);
>> -}
>> -
>> -static void
>> -ldp_fusion_destroy ()
>> -{
>> -  if (crtl->ssa->perform_pending_updates ())
>> -    cleanup_cfg (0);
>> -
>> -  free_dominance_info (CDI_DOMINATORS);
>> -
>> -  delete crtl->ssa;
>> -  crtl->ssa = nullptr;
>> -}
>> -
>>  // Given a load pair insn in PATTERN, unpack the insn, storing
>>  // the registers in REGS and returning the mem.
>>  static rtx
>> @@ -3015,16 +3202,19 @@ aarch64_destructure_store_pair (rtx regs[2], rtx pattern)
>>    return mem;
>>  }
>>  
>> -// Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx
>> -// representing the effect of writeback on the base register in WB_EFFECT,
>> -// return an insn representing a writeback variant of this pair.
>> -// LOAD_P is true iff the pair is a load.
>> -//
>> -// This is used when promoting existing non-writeback pairs to writeback
>> -// variants.
>> -static rtx
>> -aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
>> -			    bool load_p)
>> +rtx
>> +aarch64_pair_fusion::destructure_pair (rtx regs[2], rtx rti, bool load_p)
>> +{
>> +  if (load_p)
>> +    return aarch64_destructure_load_pair (regs, rti);
>> +  else
>> +    return aarch64_destructure_store_pair (regs, rti);
>> +}
>> +
>> +rtx
>> +aarch64_pair_fusion::gen_promote_writeback_pair (rtx wb_effect, rtx pair_mem,
>> +						 rtx regs[2],
>> +						 bool load_p)
>>  {
>>    auto op_mode = aarch64_operand_mode_for_pair_mode (GET_MODE (pair_mem));
>>  
>> @@ -3059,23 +3249,13 @@ aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
>>  // Given an existing pair insn INSN, look for a trailing update of
>>  // the base register which we can fold in to make this pair use
>>  // a writeback addressing mode.
>> -static void
>> -try_promote_writeback (insn_info *insn)
>> +void
>> +pair_fusion::try_promote_writeback (insn_info *insn, bool load_p)
>>  {
>> -  auto rti = insn->rtl ();
>> -  const auto attr = get_attr_ldpstp (rti);
>> -  if (attr == LDPSTP_NONE)
>> -    return;
>> -
>> -  bool load_p = (attr == LDPSTP_LDP);
>> -  gcc_checking_assert (load_p || attr == LDPSTP_STP);
>> -
>>    rtx regs[2];
>>    rtx mem = NULL_RTX;
>> -  if (load_p)
>> -    mem = aarch64_destructure_load_pair (regs, PATTERN (rti));
>> -  else
>> -    mem = aarch64_destructure_store_pair (regs, PATTERN (rti));
>> +
>> +  mem = destructure_pair (regs, PATTERN (insn->rtl ()), load_p);
> 
> Sorry for missing this before, but it's better just to have:
> 
>   rtx mem = destructure_pair (...);
> 
> i.e. declare and initialize it on the same line.
> 
Addressed in v6 of the patch.

>>    gcc_checking_assert (MEM_P (mem));
>>  
>>    poly_int64 offset;
>> @@ -3112,9 +3292,10 @@ try_promote_writeback (insn_info *insn)
>>    def_info *add_def;
>>    const insn_range_info pair_range (insn);
>>    insn_info *insns[2] = { nullptr, insn };
>> -  insn_info *trailing_add = find_trailing_add (insns, pair_range, 0, &wb_effect,
>> -					       &add_def, base_def, offset,
>> -					       access_size);
>> +  insn_info *trailing_add
>> +    = find_trailing_add (insns, pair_range, 0, &wb_effect,
>> +			 &add_def, base_def, offset,
>> +			 access_size);
>>    if (!trailing_add)
>>      return;
>>  
>> @@ -3124,8 +3305,10 @@ try_promote_writeback (insn_info *insn)
>>    insn_change del_change (trailing_add, insn_change::DELETE);
>>    insn_change *changes[] = { &pair_change, &del_change };
>>  
>> -  rtx pair_pat = aarch64_gen_writeback_pair (wb_effect, mem, regs, load_p);
>> -  validate_unshare_change (rti, &PATTERN (rti), pair_pat, true);
>> +  rtx pair_pat = gen_promote_writeback_pair (wb_effect, mem, regs,
>> +					     load_p);
> 
> The params should all fit on a single line here.
> 

Addressed in v6 of the patch.
>> +  validate_unshare_change (insn->rtl(), &PATTERN (insn->rtl()), pair_pat,
> 
> Space before '(' in function calls (two occurrences here).
> 
>> +			   true);
>>  
>>    // The pair must gain the def of the base register from the add.
>>    pair_change.new_defs = insert_access (attempt,
>> @@ -3159,14 +3342,12 @@ try_promote_writeback (insn_info *insn)
>>  // for load/store candidates.  If running after RA, also try and promote
>>  // non-writeback pairs to use writeback addressing.  Then try to fuse
>>  // candidates into pairs.
>> -void ldp_fusion_bb (bb_info *bb)
>> +void pair_fusion::ldp_fusion_bb (bb_info *bb)
>>  {
>> -  const bool track_loads
>> -    = aarch64_tune_params.ldp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
>> -  const bool track_stores
>> -    = aarch64_tune_params.stp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
>> +  const bool track_loads = track_loads_p ();
>> +  const bool track_stores = track_stores_p ();
>>  
>> -  ldp_bb_info bb_state (bb);
>> +  ldp_bb_info bb_state (bb, this);
>>  
>>    for (auto insn : bb->nondebug_insns ())
>>      {
>> @@ -3176,11 +3357,11 @@ void ldp_fusion_bb (bb_info *bb)
>>  	continue;
>>  
>>        rtx pat = PATTERN (rti);
>> +      bool load_p;
>>        if (reload_completed
>> -	  && aarch64_ldp_writeback > 1
>> -	  && GET_CODE (pat) == PARALLEL
>> -	  && XVECLEN (pat, 0) == 2)
>> -	try_promote_writeback (insn);
>> +	  && handle_writeback_opportunities (writeback::ALL)
>> +	  && pair_mem_insn_p (rti, load_p))
>> +	try_promote_writeback (insn, load_p);
>>  
>>        if (GET_CODE (pat) != SET)
>>  	continue;
>> @@ -3195,16 +3376,6 @@ void ldp_fusion_bb (bb_info *bb)
>>    bb_state.cleanup_tombstones ();
>>  }
>>  
>> -void ldp_fusion ()
>> -{
>> -  ldp_fusion_init ();
>> -
>> -  for (auto bb : crtl->ssa->bbs ())
>> -    ldp_fusion_bb (bb);
>> -
>> -  ldp_fusion_destroy ();
>> -}
>> -
>>  namespace {
>>  
>>  const pass_data pass_data_ldp_fusion =
>> @@ -3234,14 +3405,6 @@ public:
>>        if (!optimize || optimize_debug)
>>  	return false;
>>  
>> -      // If the tuning policy says never to form ldps or stps, don't run
>> -      // the pass.
>> -      if ((aarch64_tune_params.ldp_policy_model
>> -	   == AARCH64_LDP_STP_POLICY_NEVER)
>> -	  && (aarch64_tune_params.stp_policy_model
>> -	      == AARCH64_LDP_STP_POLICY_NEVER))
>> -	return false;
>> -
>>        if (reload_completed)
>>  	return flag_aarch64_late_ldp_fusion;
>>        else
>> @@ -3250,7 +3413,8 @@ public:
>>  
>>    unsigned execute (function *) final override
>>      {
>> -      ldp_fusion ();
>> +      aarch64_pair_fusion pass;
>> +      pass.run ();
>>        return 0;
>>      }
>>  };
>> -- 
>> 2.39.3
>>
> 
> Thanks,
> Alex
Thanks & Regards
Ajit
diff mbox series

Patch

diff --git a/gcc/config/aarch64/aarch64-ldp-fusion.cc b/gcc/config/aarch64/aarch64-ldp-fusion.cc
index 1d9caeab05d..e6af4b0570a 100644
--- a/gcc/config/aarch64/aarch64-ldp-fusion.cc
+++ b/gcc/config/aarch64/aarch64-ldp-fusion.cc
@@ -138,6 +138,210 @@  struct alt_base
   poly_int64 offset;
 };
 
+// Virtual base class for load/store walkers used in alias analysis.
+struct alias_walker
+{
+  virtual bool conflict_p (int &budget) const = 0;
+  virtual insn_info *insn () const = 0;
+  virtual bool valid () const = 0;
+  virtual void advance () = 0;
+};
+
+// This is used in handle_writeback_opportunities describing
+// ALL if aarch64_ldp_writeback > 1 otherwise check
+// EXISTING if aarch64_ldp_writeback.
+enum class writeback {
+  ALL,
+  EXISTING
+};
+
+struct pair_fusion {
+  pair_fusion ()
+  {
+    calculate_dominance_info (CDI_DOMINATORS);
+    df_analyze ();
+    crtl->ssa = new rtl_ssa::function_info (cfun);
+  };
+
+  // Given:
+  // - an rtx REG_OP, the non-memory operand in a load/store insn,
+  // - a machine_mode MEM_MODE, the mode of the MEM in that insn, and
+  // - a boolean LOAD_P (true iff the insn is a load), then:
+  // return true if the access should be considered an FP/SIMD access.
+  // Such accesses are segregated from GPR accesses, since we only want
+  // to form pairs for accesses that use the same register file.
+  virtual bool fpsimd_op_p (rtx, machine_mode, bool)
+  {
+    return false;
+  }
+
+  // Return true if we should consider forming ldp/stp insns from memory
+  // accesses with operand mode MODE at this stage in compilation.
+  virtual bool pair_operand_mode_ok_p (machine_mode mode) = 0;
+
+  // Return true iff REG_OP is a suitable register operand for a paired
+  // memory access, where LOAD_P is true if we're asking about loads and
+  // false for stores.  MODE gives the mode of the operand.
+  virtual bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
+				      machine_mode mode) = 0;
+
+  // Return alias check limit.
+  // This is needed to avoid unbounded quadratic behaviour when
+  // performing alias analysis.
+  virtual int pair_mem_alias_check_limit () = 0;
+
+  // Returns true if we should try to handle writeback opportunities
+  // WHICH parameter decides ALL or EXISTING writeback pairs.
+  virtual bool handle_writeback_opportunities (enum writeback which) = 0 ;
+
+  // Given BASE_MEM, the mem from the lower candidate access for a pair,
+  // and LOAD_P (true if the access is a load), check if we should proceed
+  // to form the pair given the target's code generation policy on
+  // paired accesses.
+  virtual bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) = 0;
+
+  // Generate the pattern for a paired access. PATS gives the patterns
+  // for the individual memory accesses (which by this point must share a
+  // common base register).  If WRITEBACK is non-NULL, then this rtx
+  // describes the update to the base register that should be performed by
+  // the resulting insn.  LOAD_P is true iff the accesses are loads.
+  virtual rtx gen_pair (rtx *pats, rtx writeback, bool load_p) = 0;
+
+  // Return true if memory is paired access, given INSN and LOAD_P
+  // is true for load insn and false for store insn.
+  virtual bool pair_mem_insn_p (rtx_insn *, bool &) = 0;
+
+  // Return true if we should track loads.
+  virtual bool track_loads_p ()
+  {
+    return true;
+  }
+
+  // Return true if we should track stores.
+  virtual bool track_stores_p ()
+  {
+    return true;
+  }
+
+  // Return true if OFF  is in range.
+  virtual bool pair_mem_in_range_p (HOST_WIDE_INT off) = 0;
+
+  // Given a load/store pair insn in PATTERN, unpack the insn, storing
+  // the register operands in REGS, and returning the mem. LOAD_P is true
+  // for load and false for store.
+  virtual rtx destructure_pair (rtx regs[2], rtx pattern, bool load_p) = 0;
+
+  // Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx
+  // representing the effect of writeback on the base register in WB_EFFECT,
+  // return an insn representing a writeback variant of this pair.
+  // LOAD_P is true iff the pair is a load.
+  // This is used when promoting existing non-writeback pairs to writeback
+  // variants.
+  virtual rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem,
+					  rtx regs[2], bool load_p) = 0;
+
+  void ldp_fusion_bb (bb_info *bb);
+  inline insn_info *find_trailing_add (insn_info *insns[2],
+				       const insn_range_info &pair_range,
+				       int initial_writeback,
+				       rtx *writeback_effect,
+				       def_info **add_def,
+				       def_info *base_def,
+				       poly_int64 initial_offset,
+				       unsigned access_size);
+  inline int get_viable_bases (insn_info *insns[2],
+			       vec<base_cand> &base_cands,
+			       rtx cand_mems[2],
+			       unsigned access_size,
+			       bool reversed);
+  inline void do_alias_analysis (insn_info *alias_hazards[4],
+				 alias_walker *walkers[4],
+				 bool load_p);
+  inline void try_promote_writeback (insn_info *insn, bool load_p);
+  inline void run ();
+  ~pair_fusion()
+  {
+    if (crtl->ssa->perform_pending_updates ())
+      cleanup_cfg (0);
+
+    free_dominance_info (CDI_DOMINATORS);
+
+    delete crtl->ssa;
+    crtl->ssa = nullptr;
+  }
+};
+
+// This is the main function to start the pass.
+void
+pair_fusion::run ()
+{
+  if (!track_loads_p () && !track_stores_p ())
+    return;
+
+  for (auto bb : crtl->ssa->bbs ())
+    ldp_fusion_bb (bb);
+}
+
+struct aarch64_pair_fusion : public pair_fusion
+{
+  // Before RA, we use the modes, noting that stores of constant zero
+  // operands use GPRs (even in non-integer modes).  After RA, we use
+  // the hard register numbers.
+  bool fpsimd_op_p (rtx reg_op, machine_mode mem_mode,
+		    bool load_p) override final
+  {
+    return reload_completed
+      ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
+      : (GET_MODE_CLASS (mem_mode) != MODE_INT
+	 && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
+  }
+
+  bool pair_mem_insn_p (rtx_insn *rti, bool &load_p) override final;
+  bool pair_mem_ok_with_policy (rtx base_mem, bool load_p) override final
+  {
+    return aarch64_mem_ok_with_ldpstp_policy_model (base_mem,
+						    load_p,
+						    GET_MODE (base_mem));
+  }
+  bool pair_operand_mode_ok_p (machine_mode mode);
+  rtx gen_pair (rtx *pats, rtx writeback, bool load_p);
+  bool pair_reg_operand_ok_p (bool load_p, rtx reg_op,
+			      machine_mode mode)
+  {
+    return (load_p
+	    ? aarch64_ldp_reg_operand (reg_op, mode)
+	    : aarch64_stp_reg_operand (reg_op, mode));
+  }
+  int pair_mem_alias_check_limit ()
+  {
+    return aarch64_ldp_alias_check_limit;
+  }
+  bool handle_writeback_opportunities (enum writeback which)
+  {
+    if (which == writeback::ALL)
+      return aarch64_ldp_writeback > 1;
+    else
+      return aarch64_ldp_writeback;
+  }
+  bool track_loads_p ()
+  {
+    return aarch64_tune_params.ldp_policy_model
+	   != AARCH64_LDP_STP_POLICY_NEVER;
+  }
+  bool track_stores_p ()
+  {
+    return aarch64_tune_params.stp_policy_model
+	   != AARCH64_LDP_STP_POLICY_NEVER;
+  }
+  bool pair_mem_in_range_p (HOST_WIDE_INT off)
+  {
+    return (off >= LDP_MIN_IMM && off <= LDP_MAX_IMM);
+  }
+  rtx gen_promote_writeback_pair (rtx wb_effect, rtx mem, rtx regs[2],
+				  bool load_p);
+  rtx destructure_pair (rtx regs[2], rtx rti, bool load_p);
+};
+
 // State used by the pass for a given basic block.
 struct ldp_bb_info
 {
@@ -159,9 +363,9 @@  struct ldp_bb_info
   hash_map<def_hash, alt_base> canon_base_map;
 
   static const size_t obstack_alignment = sizeof (void *);
-  bb_info *m_bb;
 
-  ldp_bb_info (bb_info *bb) : m_bb (bb), m_emitted_tombstone (false)
+  ldp_bb_info (bb_info *bb, pair_fusion *d)
+	       : m_bb (bb), m_pass (d), m_emitted_tombstone (false)
   {
     obstack_specify_allocation (&m_obstack, OBSTACK_CHUNK_SIZE,
 				obstack_alignment, obstack_chunk_alloc,
@@ -181,37 +385,75 @@  struct ldp_bb_info
   inline void track_access (insn_info *, bool load, rtx mem);
   inline void transform ();
   inline void cleanup_tombstones ();
+  inline void merge_pairs (insn_list_t &, insn_list_t &,
+		    bool load_p, unsigned access_size);
+  inline void transform_for_base (int load_size, access_group &group);
+
+  inline bool try_fuse_pair (bool load_p, unsigned access_size,
+			     insn_info *i1, insn_info *i2);
+
+  inline bool fuse_pair (bool load_p, unsigned access_size,
+		  int writeback,
+		  insn_info *i1, insn_info *i2,
+		  base_cand &base,
+		  const insn_range_info &move_range);
+
+  inline void track_tombstone (int uid);
 
 private:
   obstack m_obstack;
-
+  bb_info *m_bb;
+  pair_fusion *m_pass;
   // State for keeping track of tombstone insns emitted for this BB.
   bitmap_obstack m_bitmap_obstack;
   bitmap_head m_tombstone_bitmap;
   bool m_emitted_tombstone;
-
+  template<typename Map> inline void traverse_base_map (Map &map);
+  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
   inline splay_tree_node<access_record *> *node_alloc (access_record *);
+};
 
-  template<typename Map>
-  inline void traverse_base_map (Map &map);
-  inline void transform_for_base (int load_size, access_group &group);
-
-  inline void merge_pairs (insn_list_t &, insn_list_t &,
-			   bool load_p, unsigned access_size);
-
-  inline bool try_fuse_pair (bool load_p, unsigned access_size,
-			     insn_info *i1, insn_info *i2);
+bool
+aarch64_pair_fusion::pair_mem_insn_p (rtx_insn *rti,
+				      bool &load_p)
+{
+  rtx pat = PATTERN (rti);
+  if (GET_CODE (pat) == PARALLEL
+      && XVECLEN (pat, 0) == 2)
+    {
+      const auto attr = get_attr_ldpstp (rti);
+      if (attr == LDPSTP_NONE)
+	return false;
 
-  inline bool fuse_pair (bool load_p, unsigned access_size,
-			 int writeback,
-			 insn_info *i1, insn_info *i2,
-			 base_cand &base,
-			 const insn_range_info &move_range);
+      load_p = (attr == LDPSTP_LDP);
+      gcc_checking_assert (load_p || attr == LDPSTP_STP);
+      return true;
+    }
+  return false;
+}
 
-  inline void track_tombstone (int uid);
+rtx
+aarch64_pair_fusion::gen_pair (rtx *pats,
+			       rtx writeback,
+			       bool load_p)
+{
+  rtx pair_pat;
 
-  inline bool track_via_mem_expr (insn_info *, rtx mem, lfs_fields lfs);
-};
+  if (writeback)
+    {
+      auto patvec = gen_rtvec (3, writeback, pats[0], pats[1]);
+      return  gen_rtx_PARALLEL (VOIDmode, patvec);
+    }
+  else if (load_p)
+    return aarch64_gen_load_pair (XEXP (pats[0], 0),
+				  XEXP (pats[1], 0),
+				  XEXP (pats[0], 1));
+  else
+    return aarch64_gen_store_pair (XEXP (pats[0], 0),
+				   XEXP (pats[0], 1),
+				   XEXP (pats[1], 1));
+  return pair_pat;
+}
 
 splay_tree_node<access_record *> *
 ldp_bb_info::node_alloc (access_record *access)
@@ -312,8 +554,8 @@  any_post_modify_p (rtx x)
 
 // Return true if we should consider forming ldp/stp insns from memory
 // accesses with operand mode MODE at this stage in compilation.
-static bool
-ldp_operand_mode_ok_p (machine_mode mode)
+bool
+aarch64_pair_fusion::pair_operand_mode_ok_p (machine_mode mode)
 {
   if (!aarch64_ldpstp_operand_mode_p (mode))
     return false;
@@ -404,9 +646,10 @@  ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
   const machine_mode mem_mode = GET_MODE (mem);
   const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
 
-  // Punt on misaligned offsets.  LDP/STP instructions require offsets to be a
-  // multiple of the access size, and we believe that misaligned offsets on
-  // MEM_EXPR bases are likely to lead to misaligned offsets w.r.t. RTL bases.
+  // Punt on misaligned offsets.  Paired memory access instructions require
+  // offsets to be a multiple of the access size, and we believe that
+  // misaligned offsets on MEM_EXPR bases are likely to lead to misaligned
+  // offsets w.r.t. RTL bases.
   if (!multiple_p (offset, mem_size))
     return false;
 
@@ -430,10 +673,10 @@  ldp_bb_info::track_via_mem_expr (insn_info *insn, rtx mem, lfs_fields lfs)
 }
 
 // Main function to begin pair discovery.  Given a memory access INSN,
-// determine whether it could be a candidate for fusing into an ldp/stp,
-// and if so, track it in the appropriate data structure for this basic
-// block.  LOAD_P is true if the access is a load, and MEM is the mem
-// rtx that occurs in INSN.
+// determine whether it could be a candidate for fusing into a paired
+// access, and if so, track it in the appropriate data structure for
+// this basic block.  LOAD_P is true if the access is a load, and MEM
+// is the mem rtx that occurs in INSN.
 void
 ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
 {
@@ -441,35 +684,24 @@  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
   if (MEM_VOLATILE_P (mem))
     return;
 
-  // Ignore writeback accesses if the param says to do so.
-  if (!aarch64_ldp_writeback
+  // Ignore writeback accesses if the hook says to do so.
+  if (!m_pass->handle_writeback_opportunities (writeback::EXISTING)
       && GET_RTX_CLASS (GET_CODE (XEXP (mem, 0))) == RTX_AUTOINC)
     return;
 
   const machine_mode mem_mode = GET_MODE (mem);
-  if (!ldp_operand_mode_ok_p (mem_mode))
+  if (!m_pass->pair_operand_mode_ok_p (mem_mode))
     return;
 
   rtx reg_op = XEXP (PATTERN (insn->rtl ()), !load_p);
 
-  // Ignore the access if the register operand isn't suitable for ldp/stp.
-  if (load_p
-      ? !aarch64_ldp_reg_operand (reg_op, mem_mode)
-      : !aarch64_stp_reg_operand (reg_op, mem_mode))
+  if (!m_pass->pair_reg_operand_ok_p (load_p, reg_op, mem_mode))
     return;
 
   // We want to segregate FP/SIMD accesses from GPR accesses.
-  //
-  // Before RA, we use the modes, noting that stores of constant zero
-  // operands use GPRs (even in non-integer modes).  After RA, we use
-  // the hard register numbers.
-  const bool fpsimd_op_p
-    = reload_completed
-    ? (REG_P (reg_op) && FP_REGNUM_P (REGNO (reg_op)))
-    : (GET_MODE_CLASS (mem_mode) != MODE_INT
-       && (load_p || !aarch64_const_zero_rtx_p (reg_op)));
+  const bool fpsimd_op_p = m_pass->fpsimd_op_p (reg_op, mem_mode, load_p);
 
-  // Note ldp_operand_mode_ok_p already rejected VL modes.
+  // Note pair_operand_mode_ok_p already rejected VL modes.
   const HOST_WIDE_INT mem_size = GET_MODE_SIZE (mem_mode).to_constant ();
   const lfs_fields lfs = { load_p, fpsimd_op_p, mem_size };
 
@@ -498,8 +730,8 @@  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
   // elimination offset pre-RA, we should postpone forming pairs on such
   // accesses until after RA.
   //
-  // As it stands, addresses with offsets in range for LDR but not
-  // in range for LDP/STP are currently reloaded inefficiently,
+  // As it stands, addresses in range for an individual load/store but not
+  // for a paired access are currently reloaded inefficiently,
   // ending up with a separate base register for each pair.
   //
   // In theory LRA should make use of
@@ -510,11 +742,11 @@  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
   // constraint; this early return means we never get to the code
   // that calls targetm.legitimize_address_displacement.
   //
-  // So for now, it's better to punt when we can't be sure that the
-  // offset is in range for LDP/STP.  Out-of-range cases can then be
-  // handled after RA by the out-of-range LDP/STP peepholes.  Eventually, it
-  // would be nice to handle known out-of-range opportunities in the
-  // pass itself (for stack accesses, this would be in the post-RA pass).
+  // It's better to punt when we can't be sure that the
+  // offset is in range for paired access.  On aarch64, Out-of-range cases
+  // can then be handled after RA by the out-of-range PAIR MEM  peepholes.
+  // Eventually, it would be nice to handle known out-of-range opportunities
+  // in the pass itself (for stack accesses, this would be in the post-RA pass).
   if (!reload_completed
       && (REGNO (base) == FRAME_POINTER_REGNUM
 	  || REGNO (base) == ARG_POINTER_REGNUM))
@@ -565,8 +797,8 @@  ldp_bb_info::track_access (insn_info *insn, bool load_p, rtx mem)
 	gcc_unreachable (); // Base defs should be unique.
     }
 
-  // Punt on misaligned offsets.  LDP/STP require offsets to be a multiple of
-  // the access size.
+  // Punt on misaligned offsets.  Paired memory accesses require offsets
+  // to be a multiple of the access size.
   if (!multiple_p (mem_off, mem_size))
     return;
 
@@ -1199,8 +1431,8 @@  extract_writebacks (bool load_p, rtx pats[2], int changed)
 // base register.  If there is one, we choose the first such update after
 // PAIR_DST that is still in the same BB as our pair.  We return the new def in
 // *ADD_DEF and the resulting writeback effect in *WRITEBACK_EFFECT.
-static insn_info *
-find_trailing_add (insn_info *insns[2],
+insn_info *
+pair_fusion::find_trailing_add (insn_info *insns[2],
 		   const insn_range_info &pair_range,
 		   int initial_writeback,
 		   rtx *writeback_effect,
@@ -1278,7 +1510,7 @@  find_trailing_add (insn_info *insns[2],
 
   off_hwi /= access_size;
 
-  if (off_hwi < LDP_MIN_IMM || off_hwi > LDP_MAX_IMM)
+  if (!pair_mem_in_range_p (off_hwi))
     return nullptr;
 
   auto dump_prefix = [&]()
@@ -1792,7 +2024,7 @@  ldp_bb_info::fuse_pair (bool load_p,
     {
       if (dump_file)
 	fprintf (dump_file,
-		 "  ldp: i%d has wb but subsequent i%d has non-wb "
+		 "  load pair: i%d has wb but subsequent i%d has non-wb "
 		 "update of base (r%d), dropping wb\n",
 		 insns[0]->uid (), insns[1]->uid (), base_regno);
       gcc_assert (writeback_effect);
@@ -1815,7 +2047,7 @@  ldp_bb_info::fuse_pair (bool load_p,
     }
 
   // If either of the original insns had writeback, but the resulting pair insn
-  // does not (can happen e.g. in the ldp edge case above, or if the writeback
+  // does not (can happen e.g. in the load pair edge case above, or if the writeback
   // effects cancel out), then drop the def(s) of the base register as
   // appropriate.
   //
@@ -1834,7 +2066,7 @@  ldp_bb_info::fuse_pair (bool load_p,
   // update of the base register and try and fold it in to make this into a
   // writeback pair.
   insn_info *trailing_add = nullptr;
-  if (aarch64_ldp_writeback > 1
+  if (m_pass->handle_writeback_opportunities (writeback::ALL)
       && !writeback_effect
       && (!load_p || (!refers_to_regno_p (base_regno, base_regno + 1,
 					 XEXP (pats[0], 0), nullptr)
@@ -1842,7 +2074,7 @@  ldp_bb_info::fuse_pair (bool load_p,
 					     XEXP (pats[1], 0), nullptr))))
     {
       def_info *add_def;
-      trailing_add = find_trailing_add (insns, move_range, writeback,
+      trailing_add = m_pass->find_trailing_add (insns, move_range, writeback,
 					&writeback_effect,
 					&add_def, base.def, offsets[0],
 					access_size);
@@ -1855,35 +2087,19 @@  ldp_bb_info::fuse_pair (bool load_p,
     }
 
   // Now that we know what base mem we're going to use, check if it's OK
-  // with the ldp/stp policy.
+  // with the pair mem policy.
   rtx first_mem = XEXP (pats[0], load_p);
-  if (!aarch64_mem_ok_with_ldpstp_policy_model (first_mem,
-						load_p,
-						GET_MODE (first_mem)))
+  if (!m_pass->pair_mem_ok_with_policy (first_mem, load_p))
     {
       if (dump_file)
-	fprintf (dump_file, "punting on pair (%d,%d), ldp/stp policy says no\n",
+	fprintf (dump_file, "punting on pair (%d,%d), pair mem policy says no\n",
 		 i1->uid (), i2->uid ());
       return false;
     }
 
   rtx reg_notes = combine_reg_notes (first, second, load_p);
 
-  rtx pair_pat;
-  if (writeback_effect)
-    {
-      auto patvec = gen_rtvec (3, writeback_effect, pats[0], pats[1]);
-      pair_pat = gen_rtx_PARALLEL (VOIDmode, patvec);
-    }
-  else if (load_p)
-    pair_pat = aarch64_gen_load_pair (XEXP (pats[0], 0),
-				      XEXP (pats[1], 0),
-				      XEXP (pats[0], 1));
-  else
-    pair_pat = aarch64_gen_store_pair (XEXP (pats[0], 0),
-				       XEXP (pats[0], 1),
-				       XEXP (pats[1], 1));
-
+  rtx pair_pat = m_pass->gen_pair (pats, writeback_effect, load_p);
   insn_change *pair_change = nullptr;
   auto set_pair_pat = [pair_pat,reg_notes](insn_change *change) {
     rtx_insn *rti = change->insn ()->rtl ();
@@ -2125,15 +2341,6 @@  load_modified_by_store_p (insn_info *load,
   return false;
 }
 
-// Virtual base class for load/store walkers used in alias analysis.
-struct alias_walker
-{
-  virtual bool conflict_p (int &budget) const = 0;
-  virtual insn_info *insn () const = 0;
-  virtual bool valid () const = 0;
-  virtual void advance () = 0;
-};
-
 // Implement some common functionality used by both store_walker
 // and load_walker.
 template<bool reverse>
@@ -2251,13 +2458,13 @@  public:
 //
 // We try to maintain the invariant that if a walker becomes invalid, we
 // set its pointer to null.
-static void
-do_alias_analysis (insn_info *alias_hazards[4],
+void
+pair_fusion::do_alias_analysis (insn_info *alias_hazards[4],
 		   alias_walker *walkers[4],
 		   bool load_p)
 {
   const int n_walkers = 2 + (2 * !load_p);
-  int budget = aarch64_ldp_alias_check_limit;
+  int budget = pair_mem_alias_check_limit ();
 
   auto next_walker = [walkers,n_walkers](int current) -> int {
     for (int j = 1; j <= n_walkers; j++)
@@ -2342,8 +2549,8 @@  do_alias_analysis (insn_info *alias_hazards[4],
 //
 // Returns an integer where bit (1 << i) is set if INSNS[i] uses writeback
 // addressing.
-static int
-get_viable_bases (insn_info *insns[2],
+int
+pair_fusion::get_viable_bases (insn_info *insns[2],
 		  vec<base_cand> &base_cands,
 		  rtx cand_mems[2],
 		  unsigned access_size,
@@ -2389,7 +2596,7 @@  get_viable_bases (insn_info *insns[2],
       if (!is_lower)
 	base_off--;
 
-      if (base_off < LDP_MIN_IMM || base_off > LDP_MAX_IMM)
+      if (!pair_mem_in_range_p (base_off))
 	continue;
 
       use_info *use = find_access (insns[i]->uses (), REGNO (base));
@@ -2446,7 +2653,7 @@  get_viable_bases (insn_info *insns[2],
 }
 
 // Given two adjacent memory accesses of the same size, I1 and I2, try
-// and see if we can merge them into a ldp or stp.
+// and see if we can merge them into a paired access.
 //
 // ACCESS_SIZE gives the (common) size of a single access, LOAD_P is true
 // if the accesses are both loads, otherwise they are both stores.
@@ -2486,7 +2693,7 @@  ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
     {
       if (dump_file)
 	fprintf (dump_file,
-		 "punting on ldp due to reg conflcits (%d,%d)\n",
+		 "punting on pair mem load  due to reg conflcits (%d,%d)\n",
 		 insns[0]->uid (), insns[1]->uid ());
       return false;
     }
@@ -2504,7 +2711,7 @@  ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
 
   auto_vec<base_cand, 2> base_cands (2);
 
-  int writeback = get_viable_bases (insns, base_cands, cand_mems,
+  int writeback = m_pass->get_viable_bases (insns, base_cands, cand_mems,
 				    access_size, reversed);
   if (base_cands.is_empty ())
     {
@@ -2633,7 +2840,7 @@  ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
     walkers[1] = &backward_store_walker;
 
   if (load_p && (mem_defs[0] || mem_defs[1]))
-    do_alias_analysis (alias_hazards, walkers, load_p);
+    m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
   else
     {
       // We want to find any loads hanging off the first store.
@@ -2642,7 +2849,7 @@  ldp_bb_info::try_fuse_pair (bool load_p, unsigned access_size,
       load_walker<true> backward_load_walker (mem_defs[1], insns[1], insns[0]);
       walkers[2] = &forward_load_walker;
       walkers[3] = &backward_load_walker;
-      do_alias_analysis (alias_hazards, walkers, load_p);
+      m_pass->do_alias_analysis (alias_hazards, walkers, load_p);
       // Now consolidate hazards back down.
       if (alias_hazards[2]
 	  && (!alias_hazards[0] || (*alias_hazards[2] < *alias_hazards[0])))
@@ -2956,26 +3163,6 @@  ldp_bb_info::transform ()
   traverse_base_map (def_map);
 }
 
-static void
-ldp_fusion_init ()
-{
-  calculate_dominance_info (CDI_DOMINATORS);
-  df_analyze ();
-  crtl->ssa = new rtl_ssa::function_info (cfun);
-}
-
-static void
-ldp_fusion_destroy ()
-{
-  if (crtl->ssa->perform_pending_updates ())
-    cleanup_cfg (0);
-
-  free_dominance_info (CDI_DOMINATORS);
-
-  delete crtl->ssa;
-  crtl->ssa = nullptr;
-}
-
 // Given a load pair insn in PATTERN, unpack the insn, storing
 // the registers in REGS and returning the mem.
 static rtx
@@ -3015,16 +3202,19 @@  aarch64_destructure_store_pair (rtx regs[2], rtx pattern)
   return mem;
 }
 
-// Given a pair mem in PAIR_MEM, register operands in REGS, and an rtx
-// representing the effect of writeback on the base register in WB_EFFECT,
-// return an insn representing a writeback variant of this pair.
-// LOAD_P is true iff the pair is a load.
-//
-// This is used when promoting existing non-writeback pairs to writeback
-// variants.
-static rtx
-aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
-			    bool load_p)
+rtx
+aarch64_pair_fusion::destructure_pair (rtx regs[2], rtx rti, bool load_p)
+{
+  if (load_p)
+    return aarch64_destructure_load_pair (regs, rti);
+  else
+    return aarch64_destructure_store_pair (regs, rti);
+}
+
+rtx
+aarch64_pair_fusion::gen_promote_writeback_pair (rtx wb_effect, rtx pair_mem,
+						 rtx regs[2],
+						 bool load_p)
 {
   auto op_mode = aarch64_operand_mode_for_pair_mode (GET_MODE (pair_mem));
 
@@ -3059,23 +3249,13 @@  aarch64_gen_writeback_pair (rtx wb_effect, rtx pair_mem, rtx regs[2],
 // Given an existing pair insn INSN, look for a trailing update of
 // the base register which we can fold in to make this pair use
 // a writeback addressing mode.
-static void
-try_promote_writeback (insn_info *insn)
+void
+pair_fusion::try_promote_writeback (insn_info *insn, bool load_p)
 {
-  auto rti = insn->rtl ();
-  const auto attr = get_attr_ldpstp (rti);
-  if (attr == LDPSTP_NONE)
-    return;
-
-  bool load_p = (attr == LDPSTP_LDP);
-  gcc_checking_assert (load_p || attr == LDPSTP_STP);
-
   rtx regs[2];
   rtx mem = NULL_RTX;
-  if (load_p)
-    mem = aarch64_destructure_load_pair (regs, PATTERN (rti));
-  else
-    mem = aarch64_destructure_store_pair (regs, PATTERN (rti));
+
+  mem = destructure_pair (regs, PATTERN (insn->rtl ()), load_p);
   gcc_checking_assert (MEM_P (mem));
 
   poly_int64 offset;
@@ -3112,9 +3292,10 @@  try_promote_writeback (insn_info *insn)
   def_info *add_def;
   const insn_range_info pair_range (insn);
   insn_info *insns[2] = { nullptr, insn };
-  insn_info *trailing_add = find_trailing_add (insns, pair_range, 0, &wb_effect,
-					       &add_def, base_def, offset,
-					       access_size);
+  insn_info *trailing_add
+    = find_trailing_add (insns, pair_range, 0, &wb_effect,
+			 &add_def, base_def, offset,
+			 access_size);
   if (!trailing_add)
     return;
 
@@ -3124,8 +3305,10 @@  try_promote_writeback (insn_info *insn)
   insn_change del_change (trailing_add, insn_change::DELETE);
   insn_change *changes[] = { &pair_change, &del_change };
 
-  rtx pair_pat = aarch64_gen_writeback_pair (wb_effect, mem, regs, load_p);
-  validate_unshare_change (rti, &PATTERN (rti), pair_pat, true);
+  rtx pair_pat = gen_promote_writeback_pair (wb_effect, mem, regs,
+					     load_p);
+  validate_unshare_change (insn->rtl(), &PATTERN (insn->rtl()), pair_pat,
+			   true);
 
   // The pair must gain the def of the base register from the add.
   pair_change.new_defs = insert_access (attempt,
@@ -3159,14 +3342,12 @@  try_promote_writeback (insn_info *insn)
 // for load/store candidates.  If running after RA, also try and promote
 // non-writeback pairs to use writeback addressing.  Then try to fuse
 // candidates into pairs.
-void ldp_fusion_bb (bb_info *bb)
+void pair_fusion::ldp_fusion_bb (bb_info *bb)
 {
-  const bool track_loads
-    = aarch64_tune_params.ldp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
-  const bool track_stores
-    = aarch64_tune_params.stp_policy_model != AARCH64_LDP_STP_POLICY_NEVER;
+  const bool track_loads = track_loads_p ();
+  const bool track_stores = track_stores_p ();
 
-  ldp_bb_info bb_state (bb);
+  ldp_bb_info bb_state (bb, this);
 
   for (auto insn : bb->nondebug_insns ())
     {
@@ -3176,11 +3357,11 @@  void ldp_fusion_bb (bb_info *bb)
 	continue;
 
       rtx pat = PATTERN (rti);
+      bool load_p;
       if (reload_completed
-	  && aarch64_ldp_writeback > 1
-	  && GET_CODE (pat) == PARALLEL
-	  && XVECLEN (pat, 0) == 2)
-	try_promote_writeback (insn);
+	  && handle_writeback_opportunities (writeback::ALL)
+	  && pair_mem_insn_p (rti, load_p))
+	try_promote_writeback (insn, load_p);
 
       if (GET_CODE (pat) != SET)
 	continue;
@@ -3195,16 +3376,6 @@  void ldp_fusion_bb (bb_info *bb)
   bb_state.cleanup_tombstones ();
 }
 
-void ldp_fusion ()
-{
-  ldp_fusion_init ();
-
-  for (auto bb : crtl->ssa->bbs ())
-    ldp_fusion_bb (bb);
-
-  ldp_fusion_destroy ();
-}
-
 namespace {
 
 const pass_data pass_data_ldp_fusion =
@@ -3234,14 +3405,6 @@  public:
       if (!optimize || optimize_debug)
 	return false;
 
-      // If the tuning policy says never to form ldps or stps, don't run
-      // the pass.
-      if ((aarch64_tune_params.ldp_policy_model
-	   == AARCH64_LDP_STP_POLICY_NEVER)
-	  && (aarch64_tune_params.stp_policy_model
-	      == AARCH64_LDP_STP_POLICY_NEVER))
-	return false;
-
       if (reload_completed)
 	return flag_aarch64_late_ldp_fusion;
       else
@@ -3250,7 +3413,8 @@  public:
 
   unsigned execute (function *) final override
     {
-      ldp_fusion ();
+      aarch64_pair_fusion pass;
+      pass.run ();
       return 0;
     }
 };