Add support for in-order addition reduction using SVE FADDA

[thirdparty/gcc.git] / gcc / optabs-query.c

/* IR-agnostic target query functions relating to optabs
   Copyright (C) 1987-2018 Free Software Foundation, Inc.

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.

GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3.  If not see
<http://www.gnu.org/licenses/>.  */


#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "target.h"
#include "insn-codes.h"
#include "optabs-query.h"
#include "optabs-libfuncs.h"
#include "insn-config.h"
#include "rtl.h"
#include "recog.h"
#include "vec-perm-indices.h"

struct target_optabs default_target_optabs;
struct target_optabs *this_fn_optabs = &default_target_optabs;
#if SWITCHABLE_TARGET
struct target_optabs *this_target_optabs = &default_target_optabs;
#endif

/* Return the insn used to perform conversion OP from mode FROM_MODE
   to mode TO_MODE; return CODE_FOR_nothing if the target does not have
   such an insn, or if it is unsuitable for optimization type OPT_TYPE.  */

insn_code
convert_optab_handler (convert_optab optab, machine_mode to_mode,
		       machine_mode from_mode, optimization_type opt_type)
{
  insn_code icode = convert_optab_handler (optab, to_mode, from_mode);
  if (icode == CODE_FOR_nothing
      || !targetm.optab_supported_p (optab, to_mode, from_mode, opt_type))
    return CODE_FOR_nothing;
  return icode;
}

/* Return the insn used to implement mode MODE of OP; return
   CODE_FOR_nothing if the target does not have such an insn,
   or if it is unsuitable for optimization type OPT_TYPE.  */

insn_code
direct_optab_handler (convert_optab optab, machine_mode mode,
		      optimization_type opt_type)
{
  insn_code icode = direct_optab_handler (optab, mode);
  if (icode == CODE_FOR_nothing
      || !targetm.optab_supported_p (optab, mode, mode, opt_type))
    return CODE_FOR_nothing;
  return icode;
}

/* Enumerates the possible types of structure operand to an
   extraction_insn.  */
enum extraction_type { ET_unaligned_mem, ET_reg };

/* Check whether insv, extv or extzv pattern ICODE can be used for an
   insertion or extraction of type TYPE on a structure of mode MODE.
   Return true if so and fill in *INSN accordingly.  STRUCT_OP is the
   operand number of the structure (the first sign_extract or zero_extract
   operand) and FIELD_OP is the operand number of the field (the other
   side of the set from the sign_extract or zero_extract).  */

static bool
get_traditional_extraction_insn (extraction_insn *insn,
				 enum extraction_type type,
				 machine_mode mode,
				 enum insn_code icode,
				 int struct_op, int field_op)
{
  const struct insn_data_d *data = &insn_data[icode];

  machine_mode struct_mode = data->operand[struct_op].mode;
  if (struct_mode == VOIDmode)
    struct_mode = word_mode;
  if (mode != struct_mode)
    return false;

  machine_mode field_mode = data->operand[field_op].mode;
  if (field_mode == VOIDmode)
    field_mode = word_mode;

  machine_mode pos_mode = data->operand[struct_op + 2].mode;
  if (pos_mode == VOIDmode)
    pos_mode = word_mode;

  insn->icode = icode;
  insn->field_mode = as_a <scalar_int_mode> (field_mode);
  if (type == ET_unaligned_mem)
    insn->struct_mode = byte_mode;
  else if (struct_mode == BLKmode)
    insn->struct_mode = opt_scalar_int_mode ();
  else
    insn->struct_mode = as_a <scalar_int_mode> (struct_mode);
  insn->pos_mode = as_a <scalar_int_mode> (pos_mode);
  return true;
}

/* Return true if an optab exists to perform an insertion or extraction
   of type TYPE in mode MODE.  Describe the instruction in *INSN if so.

   REG_OPTAB is the optab to use for register structures and
   MISALIGN_OPTAB is the optab to use for misaligned memory structures.
   POS_OP is the operand number of the bit position.  */

static bool
get_optab_extraction_insn (struct extraction_insn *insn,
			   enum extraction_type type,
			   machine_mode mode, direct_optab reg_optab,
			   direct_optab misalign_optab, int pos_op)
{
  direct_optab optab = (type == ET_unaligned_mem ? misalign_optab : reg_optab);
  enum insn_code icode = direct_optab_handler (optab, mode);
  if (icode == CODE_FOR_nothing)
    return false;

  const struct insn_data_d *data = &insn_data[icode];

  machine_mode pos_mode = data->operand[pos_op].mode;
  if (pos_mode == VOIDmode)
    pos_mode = word_mode;

  insn->icode = icode;
  insn->field_mode = as_a <scalar_int_mode> (mode);
  if (type == ET_unaligned_mem)
    insn->struct_mode = opt_scalar_int_mode ();
  else
    insn->struct_mode = insn->field_mode;
  insn->pos_mode = as_a <scalar_int_mode> (pos_mode);
  return true;
}

/* Return true if an instruction exists to perform an insertion or
   extraction (PATTERN says which) of type TYPE in mode MODE.
   Describe the instruction in *INSN if so.  */

static bool
get_extraction_insn (extraction_insn *insn,
		     enum extraction_pattern pattern,
		     enum extraction_type type,
		     machine_mode mode)
{
  switch (pattern)
    {
    case EP_insv:
      if (targetm.have_insv ()
	  && get_traditional_extraction_insn (insn, type, mode,
					      targetm.code_for_insv, 0, 3))
	return true;
      return get_optab_extraction_insn (insn, type, mode, insv_optab,
					insvmisalign_optab, 2);

    case EP_extv:
      if (targetm.have_extv ()
	  && get_traditional_extraction_insn (insn, type, mode,
					      targetm.code_for_extv, 1, 0))
	return true;
      return get_optab_extraction_insn (insn, type, mode, extv_optab,
					extvmisalign_optab, 3);

    case EP_extzv:
      if (targetm.have_extzv ()
	  && get_traditional_extraction_insn (insn, type, mode,
					      targetm.code_for_extzv, 1, 0))
	return true;
      return get_optab_extraction_insn (insn, type, mode, extzv_optab,
					extzvmisalign_optab, 3);

    default:
      gcc_unreachable ();
    }
}

/* Return true if an instruction exists to access a field of mode
   FIELDMODE in a structure that has STRUCT_BITS significant bits.
   Describe the "best" such instruction in *INSN if so.  PATTERN and
   TYPE describe the type of insertion or extraction we want to perform.

   For an insertion, the number of significant structure bits includes
   all bits of the target.  For an extraction, it need only include the
   most significant bit of the field.  Larger widths are acceptable
   in both cases.  */

static bool
get_best_extraction_insn (extraction_insn *insn,
			  enum extraction_pattern pattern,
			  enum extraction_type type,
			  unsigned HOST_WIDE_INT struct_bits,
			  machine_mode field_mode)
{
  opt_scalar_int_mode mode_iter;
  FOR_EACH_MODE_FROM (mode_iter, smallest_int_mode_for_size (struct_bits))
    {
      scalar_int_mode mode = mode_iter.require ();
      if (get_extraction_insn (insn, pattern, type, mode))
	{
	  FOR_EACH_MODE_FROM (mode_iter, mode)
	    {
	      mode = mode_iter.require ();
	      if (maybe_gt (GET_MODE_SIZE (mode), GET_MODE_SIZE (field_mode))
		  || TRULY_NOOP_TRUNCATION_MODES_P (insn->field_mode,
						    field_mode))
		break;
	      get_extraction_insn (insn, pattern, type, mode);
	    }
	  return true;
	}
    }
  return false;
}

/* Return true if an instruction exists to access a field of mode
   FIELDMODE in a register structure that has STRUCT_BITS significant bits.
   Describe the "best" such instruction in *INSN if so.  PATTERN describes
   the type of insertion or extraction we want to perform.

   For an insertion, the number of significant structure bits includes
   all bits of the target.  For an extraction, it need only include the
   most significant bit of the field.  Larger widths are acceptable
   in both cases.  */

bool
get_best_reg_extraction_insn (extraction_insn *insn,
			      enum extraction_pattern pattern,
			      unsigned HOST_WIDE_INT struct_bits,
			      machine_mode field_mode)
{
  return get_best_extraction_insn (insn, pattern, ET_reg, struct_bits,
				   field_mode);
}

/* Return true if an instruction exists to access a field of BITSIZE
   bits starting BITNUM bits into a memory structure.  Describe the
   "best" such instruction in *INSN if so.  PATTERN describes the type
   of insertion or extraction we want to perform and FIELDMODE is the
   natural mode of the extracted field.

   The instructions considered here only access bytes that overlap
   the bitfield; they do not touch any surrounding bytes.  */

bool
get_best_mem_extraction_insn (extraction_insn *insn,
			      enum extraction_pattern pattern,
			      HOST_WIDE_INT bitsize, HOST_WIDE_INT bitnum,
			      machine_mode field_mode)
{
  unsigned HOST_WIDE_INT struct_bits = (bitnum % BITS_PER_UNIT
					+ bitsize
					+ BITS_PER_UNIT - 1);
  struct_bits -= struct_bits % BITS_PER_UNIT;
  return get_best_extraction_insn (insn, pattern, ET_unaligned_mem,
				   struct_bits, field_mode);
}

/* Return the insn code used to extend FROM_MODE to TO_MODE.
   UNSIGNEDP specifies zero-extension instead of sign-extension.  If
   no such operation exists, CODE_FOR_nothing will be returned.  */

enum insn_code
can_extend_p (machine_mode to_mode, machine_mode from_mode,
	      int unsignedp)
{
  if (unsignedp < 0 && targetm.have_ptr_extend ())
    return targetm.code_for_ptr_extend;

  convert_optab tab = unsignedp ? zext_optab : sext_optab;
  return convert_optab_handler (tab, to_mode, from_mode);
}

/* Return the insn code to convert fixed-point mode FIXMODE to floating-point
   mode FLTMODE, or CODE_FOR_nothing if no such instruction exists.
   UNSIGNEDP specifies whether FIXMODE is unsigned.  */

enum insn_code
can_float_p (machine_mode fltmode, machine_mode fixmode,
	     int unsignedp)
{
  convert_optab tab = unsignedp ? ufloat_optab : sfloat_optab;
  return convert_optab_handler (tab, fltmode, fixmode);
}

/* Return the insn code to convert floating-point mode FLTMODE to fixed-point
   mode FIXMODE, or CODE_FOR_nothing if no such instruction exists.
   UNSIGNEDP specifies whether FIXMODE is unsigned.

   On a successful return, set *TRUNCP_PTR to true if it is necessary to
   output an explicit FTRUNC before the instruction.  */

enum insn_code
can_fix_p (machine_mode fixmode, machine_mode fltmode,
	   int unsignedp, bool *truncp_ptr)
{
  convert_optab tab;
  enum insn_code icode;

  tab = unsignedp ? ufixtrunc_optab : sfixtrunc_optab;
  icode = convert_optab_handler (tab, fixmode, fltmode);
  if (icode != CODE_FOR_nothing)
    {
      *truncp_ptr = false;
      return icode;
    }

  /* FIXME: This requires a port to define both FIX and FTRUNC pattern
     for this to work.  We need to rework the fix* and ftrunc* patterns
     and documentation.  */
  tab = unsignedp ? ufix_optab : sfix_optab;
  icode = convert_optab_handler (tab, fixmode, fltmode);
  if (icode != CODE_FOR_nothing
      && optab_handler (ftrunc_optab, fltmode) != CODE_FOR_nothing)
    {
      *truncp_ptr = true;
      return icode;
    }

  return CODE_FOR_nothing;
}

/* Return nonzero if a conditional move of mode MODE is supported.

   This function is for combine so it can tell whether an insn that looks
   like a conditional move is actually supported by the hardware.  If we
   guess wrong we lose a bit on optimization, but that's it.  */
/* ??? sparc64 supports conditionally moving integers values based on fp
   comparisons, and vice versa.  How do we handle them?  */

bool
can_conditionally_move_p (machine_mode mode)
{
  return direct_optab_handler (movcc_optab, mode) != CODE_FOR_nothing;
}

/* If a target doesn't implement a permute on a vector with multibyte
   elements, we can try to do the same permute on byte elements.
   If this makes sense for vector mode MODE then return the appropriate
   byte vector mode.  */

opt_machine_mode
qimode_for_vec_perm (machine_mode mode)
{
  machine_mode qimode;
  if (GET_MODE_INNER (mode) != QImode
      && mode_for_vector (QImode, GET_MODE_SIZE (mode)).exists (&qimode)
      && VECTOR_MODE_P (qimode))
    return qimode;
  return opt_machine_mode ();
}

/* Return true if selector SEL can be represented in the integer
   equivalent of vector mode MODE.  */

bool
selector_fits_mode_p (machine_mode mode, const vec_perm_indices &sel)
{
  unsigned HOST_WIDE_INT mask = GET_MODE_MASK (GET_MODE_INNER (mode));
  return (mask == HOST_WIDE_INT_M1U
	  || sel.all_in_range_p (0, mask + 1));
}

/* Return true if VEC_PERM_EXPRs with variable selector operands can be
   expanded using SIMD extensions of the CPU.  MODE is the mode of the
   vectors being permuted.  */

bool
can_vec_perm_var_p (machine_mode mode)
{
  /* If the target doesn't implement a vector mode for the vector type,
     then no operations are supported.  */
  if (!VECTOR_MODE_P (mode))
    return false;

  if (direct_optab_handler (vec_perm_optab, mode) != CODE_FOR_nothing)
    return true;

  /* We allow fallback to a QI vector mode, and adjust the mask.  */
  machine_mode qimode;
  if (!qimode_for_vec_perm (mode).exists (&qimode)
      || maybe_gt (GET_MODE_NUNITS (qimode), GET_MODE_MASK (QImode) + 1))
    return false;

  if (direct_optab_handler (vec_perm_optab, qimode) == CODE_FOR_nothing)
    return false;

  /* In order to support the lowering of variable permutations,
     we need to support shifts and adds.  */
  if (GET_MODE_UNIT_SIZE (mode) > 2
      && optab_handler (ashl_optab, mode) == CODE_FOR_nothing
      && optab_handler (vashl_optab, mode) == CODE_FOR_nothing)
    return false;
  if (optab_handler (add_optab, qimode) == CODE_FOR_nothing)
    return false;

  return true;
}

/* Return true if the target directly supports VEC_PERM_EXPRs on vectors
   of mode MODE using the selector SEL.  ALLOW_VARIABLE_P is true if it
   is acceptable to force the selector into a register and use a variable
   permute (if the target supports that).

   Note that additional permutations representing whole-vector shifts may
   also be handled via the vec_shr optab, but only where the second input
   vector is entirely constant zeroes; this case is not dealt with here.  */

bool
can_vec_perm_const_p (machine_mode mode, const vec_perm_indices &sel,
		      bool allow_variable_p)
{
  /* If the target doesn't implement a vector mode for the vector type,
     then no operations are supported.  */
  if (!VECTOR_MODE_P (mode))
    return false;

  /* It's probably cheaper to test for the variable case first.  */
  if (allow_variable_p && selector_fits_mode_p (mode, sel))
    {
      if (direct_optab_handler (vec_perm_optab, mode) != CODE_FOR_nothing)
	return true;

      /* Unlike can_vec_perm_var_p, we don't need to test for optabs
	 related computing the QImode selector, since that happens at
	 compile time.  */
      machine_mode qimode;
      if (qimode_for_vec_perm (mode).exists (&qimode))
	{
	  vec_perm_indices qimode_indices;
	  qimode_indices.new_expanded_vector (sel, GET_MODE_UNIT_SIZE (mode));
	  if (selector_fits_mode_p (qimode, qimode_indices)
	      && (direct_optab_handler (vec_perm_optab, qimode)
		  != CODE_FOR_nothing))
	    return true;
	}
    }

  if (targetm.vectorize.vec_perm_const != NULL)
    {
      if (targetm.vectorize.vec_perm_const (mode, NULL_RTX, NULL_RTX,
					    NULL_RTX, sel))
	return true;

      /* ??? For completeness, we ought to check the QImode version of
	 vec_perm_const_optab.  But all users of this implicit lowering
	 feature implement the variable vec_perm_optab, and the ia64
	 port specifically doesn't want us to lower V2SF operations
	 into integer operations.  */
    }

  return false;
}

/* Find a widening optab even if it doesn't widen as much as we want.
   E.g. if from_mode is HImode, and to_mode is DImode, and there is no
   direct HI->SI insn, then return SI->DI, if that exists.  */

enum insn_code
find_widening_optab_handler_and_mode (optab op, machine_mode to_mode,
				      machine_mode from_mode,
				      machine_mode *found_mode)
{
  gcc_checking_assert (GET_MODE_CLASS (from_mode) == GET_MODE_CLASS (to_mode));
  gcc_checking_assert (from_mode < to_mode);
  FOR_EACH_MODE (from_mode, from_mode, to_mode)
    {
      enum insn_code handler = convert_optab_handler (op, to_mode, from_mode);

      if (handler != CODE_FOR_nothing)
	{
	  if (found_mode)
	    *found_mode = from_mode;
	  return handler;
	}
    }

  return CODE_FOR_nothing;
}

/* Return non-zero if a highpart multiply is supported of can be synthisized.
   For the benefit of expand_mult_highpart, the return value is 1 for direct,
   2 for even/odd widening, and 3 for hi/lo widening.  */

int
can_mult_highpart_p (machine_mode mode, bool uns_p)
{
  optab op;

  op = uns_p ? umul_highpart_optab : smul_highpart_optab;
  if (optab_handler (op, mode) != CODE_FOR_nothing)
    return 1;

  /* If the mode is an integral vector, synth from widening operations.  */
  if (GET_MODE_CLASS (mode) != MODE_VECTOR_INT)
    return 0;

  poly_int64 nunits = GET_MODE_NUNITS (mode);

  op = uns_p ? vec_widen_umult_even_optab : vec_widen_smult_even_optab;
  if (optab_handler (op, mode) != CODE_FOR_nothing)
    {
      op = uns_p ? vec_widen_umult_odd_optab : vec_widen_smult_odd_optab;
      if (optab_handler (op, mode) != CODE_FOR_nothing)
	{
	  /* The encoding has 2 interleaved stepped patterns.  */
	  vec_perm_builder sel (nunits, 2, 3);
	  for (unsigned int i = 0; i < 6; ++i)
	    sel.quick_push (!BYTES_BIG_ENDIAN
			    + (i & ~1)
			    + ((i & 1) ? nunits : 0));
	  vec_perm_indices indices (sel, 2, nunits);
	  if (can_vec_perm_const_p (mode, indices))
	    return 2;
	}
    }

  op = uns_p ? vec_widen_umult_hi_optab : vec_widen_smult_hi_optab;
  if (optab_handler (op, mode) != CODE_FOR_nothing)
    {
      op = uns_p ? vec_widen_umult_lo_optab : vec_widen_smult_lo_optab;
      if (optab_handler (op, mode) != CODE_FOR_nothing)
	{
	  /* The encoding has a single stepped pattern.  */
	  vec_perm_builder sel (nunits, 1, 3);
	  for (unsigned int i = 0; i < 3; ++i)
	    sel.quick_push (2 * i + (BYTES_BIG_ENDIAN ? 0 : 1));
	  vec_perm_indices indices (sel, 2, nunits);
	  if (can_vec_perm_const_p (mode, indices))
	    return 3;
	}
    }

  return 0;
}

/* Return true if target supports vector masked load/store for mode.  */

bool
can_vec_mask_load_store_p (machine_mode mode,
			   machine_mode mask_mode,
			   bool is_load)
{
  optab op = is_load ? maskload_optab : maskstore_optab;
  machine_mode vmode;

  /* If mode is vector mode, check it directly.  */
  if (VECTOR_MODE_P (mode))
    return convert_optab_handler (op, mode, mask_mode) != CODE_FOR_nothing;

  /* Otherwise, return true if there is some vector mode with
     the mask load/store supported.  */

  /* See if there is any chance the mask load or store might be
     vectorized.  If not, punt.  */
  scalar_mode smode;
  if (!is_a <scalar_mode> (mode, &smode))
    return false;

  vmode = targetm.vectorize.preferred_simd_mode (smode);
  if (!VECTOR_MODE_P (vmode))
    return false;

  if ((targetm.vectorize.get_mask_mode
       (GET_MODE_NUNITS (vmode), GET_MODE_SIZE (vmode)).exists (&mask_mode))
      && convert_optab_handler (op, vmode, mask_mode) != CODE_FOR_nothing)
    return true;

  auto_vector_sizes vector_sizes;
  targetm.vectorize.autovectorize_vector_sizes (&vector_sizes);
  for (unsigned int i = 0; i < vector_sizes.length (); ++i)
    {
      poly_uint64 cur = vector_sizes[i];
      poly_uint64 nunits;
      if (!multiple_p (cur, GET_MODE_SIZE (smode), &nunits))
	continue;
      if (mode_for_vector (smode, nunits).exists (&vmode)
	  && VECTOR_MODE_P (vmode)
	  && targetm.vectorize.get_mask_mode (nunits, cur).exists (&mask_mode)
	  && convert_optab_handler (op, vmode, mask_mode) != CODE_FOR_nothing)
	return true;
    }
  return false;
}

/* Return true if there is a compare_and_swap pattern.  */

bool
can_compare_and_swap_p (machine_mode mode, bool allow_libcall)
{
  enum insn_code icode;

  /* Check for __atomic_compare_and_swap.  */
  icode = direct_optab_handler (atomic_compare_and_swap_optab, mode);
  if (icode != CODE_FOR_nothing)
    return true;

  /* Check for __sync_compare_and_swap.  */
  icode = optab_handler (sync_compare_and_swap_optab, mode);
  if (icode != CODE_FOR_nothing)
    return true;
  if (allow_libcall && optab_libfunc (sync_compare_and_swap_optab, mode))
    return true;

  /* No inline compare and swap.  */
  return false;
}

/* Return true if an atomic exchange can be performed.  */

bool
can_atomic_exchange_p (machine_mode mode, bool allow_libcall)
{
  enum insn_code icode;

  /* Check for __atomic_exchange.  */
  icode = direct_optab_handler (atomic_exchange_optab, mode);
  if (icode != CODE_FOR_nothing)
    return true;

  /* Don't check __sync_test_and_set, as on some platforms that
     has reduced functionality.  Targets that really do support
     a proper exchange should simply be updated to the __atomics.  */

  return can_compare_and_swap_p (mode, allow_libcall);
}

/* Return true if an atomic load can be performed without falling back to
   a compare-and-swap.  */

bool
can_atomic_load_p (machine_mode mode)
{
  enum insn_code icode;

  /* Does the target supports the load directly?  */
  icode = direct_optab_handler (atomic_load_optab, mode);
  if (icode != CODE_FOR_nothing)
    return true;

  /* If the size of the object is greater than word size on this target,
     then we assume that a load will not be atomic.  Also see
     expand_atomic_load.  */
  return known_le (GET_MODE_PRECISION (mode), BITS_PER_WORD);
}

/* Determine whether "1 << x" is relatively cheap in word_mode.  */

bool
lshift_cheap_p (bool speed_p)
{
  /* FIXME: This should be made target dependent via this "this_target"
     mechanism, similar to e.g. can_copy_init_p in gcse.c.  */
  static bool init[2] = { false, false };
  static bool cheap[2] = { true, true };

  /* If the targer has no lshift in word_mode, the operation will most
     probably not be cheap.  ??? Does GCC even work for such targets?  */
  if (optab_handler (ashl_optab, word_mode) == CODE_FOR_nothing)
    return false;

  if (!init[speed_p])
    {
      rtx reg = gen_raw_REG (word_mode, 10000);
      int cost = set_src_cost (gen_rtx_ASHIFT (word_mode, const1_rtx, reg),
			       word_mode, speed_p);
      cheap[speed_p] = cost < COSTS_N_INSNS (3);
      init[speed_p] = true;
    }

  return cheap[speed_p];
}