2014-05-23 Thomas Preud'homme <thomas.preudhomme@arm.com>

	PR tree-optimization/54733
gcc/
	* tree-ssa-math-opts.c (nop_stats): New "bswap_stats" structure.
	(CMPNOP): Define.
	(find_bswap_or_nop_load): New.
	(find_bswap_1): Renamed to ...
	(find_bswap_or_nop_1): This. Also add support for memory source.
	(find_bswap): Renamed to ...
	(find_bswap_or_nop): This. Also add support for memory source and
	detection of bitwise operations equivalent to load in host endianness.
	(execute_optimize_bswap): Likewise. Also move its leading comment back
	in place and split statement transformation into ...
	(bswap_replace): This.

gcc/testsuite
	* gcc.dg/optimize-bswapdi-3.c: New test to check extension of bswap
	optimization to support memory sources and bitwise operations
	equivalent to load in host endianness.
	* gcc.dg/optimize-bswaphi-1.c: Likewise.
	* gcc.dg/optimize-bswapsi-2.c: Likewise.
	* gcc.c-torture/execute/bswap-2.c: Likewise.


git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@210843 138bc75d-0d04-0410-961f-82ee72b054a4

1 files changed, 405 insertions, 110 deletions

diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
index 336626d0886..7c8c63ac99a 100644
--- a/gcc/tree-ssa-math-opts.c
+++ b/gcc/tree-ssa-math-opts.c
@@ -98,6 +98,7 @@ along with GCC; see the file COPYING3.  If not see
 #include "is-a.h"
 #include "gimple.h"
 #include "gimple-iterator.h"
+#include "gimplify.h"
 #include "gimplify-me.h"
 #include "stor-layout.h"
 #include "gimple-ssa.h"
@@ -170,15 +171,15 @@ static struct
 
 static struct
 {
-  /* Number of hand-written 16-bit bswaps found.  */
+  /* Number of hand-written 16-bit nop / bswaps found.  */
   int found_16bit;
 
-  /* Number of hand-written 32-bit bswaps found.  */
+  /* Number of hand-written 32-bit nop / bswaps found.  */
   int found_32bit;
 
-  /* Number of hand-written 64-bit bswaps found.  */
+  /* Number of hand-written 64-bit nop / bswaps found.  */
   int found_64bit;
-} bswap_stats;
+} nop_stats, bswap_stats;
 
 static struct
 {
@@ -1604,13 +1605,43 @@ make_pass_cse_sincos (gcc::context *ctxt)
 
    0    - byte has the value 0
    1..size - byte contains the content of the byte
-   number indexed with that value minus one  */
+   number indexed with that value minus one.
+
+   To detect permutations on memory sources (arrays and structures), a symbolic
+   number is also associated a base address (the array or structure the load is
+   made from), an offset from the base address and a range which gives the
+   difference between the highest and lowest accessed memory location to make
+   such a symbolic number. The range is thus different from size which reflects
+   the size of the type of current expression. Note that for non memory source,
+   range holds the same value as size.
+
+   For instance, for an array char a[], (short) a[0] | (short) a[3] would have
+   a size of 2 but a range of 4 while (short) a[0] | ((short) a[0] << 1) would
+   still have a size of 2 but this time a range of 1.  */
 
 struct symbolic_number {
   unsigned HOST_WIDEST_INT n;
   int size;
+  tree base_addr;
+  tree offset;
+  HOST_WIDE_INT bytepos;
+  tree alias_set;
+  tree vuse;
+  unsigned HOST_WIDE_INT range;
 };
 
+/* The number which the find_bswap_or_nop_1 result should match in
+   order to have a nop.  The number is masked according to the size of
+   the symbolic number before using it.  */
+#define CMPNOP (sizeof (HOST_WIDEST_INT) < 8 ? 0 : \
+  (unsigned HOST_WIDEST_INT)0x08070605 << 32 | 0x04030201)
+
+/* The number which the find_bswap_or_nop_1 result should match in
+   order to have a byte swap.  The number is masked according to the
+   size of the symbolic number before using it.  */
+#define CMPXCHG (sizeof (HOST_WIDEST_INT) < 8 ? 0 : \
+  (unsigned HOST_WIDEST_INT)0x01020304 << 32 | 0x05060708)
+
 /* Perform a SHIFT or ROTATE operation by COUNT bits on symbolic
    number N.  Return false if the requested operation is not permitted
    on a symbolic number.  */
@@ -1670,13 +1701,76 @@ verify_symbolic_number_p (struct symbolic_number *n, gimple stmt)
   return true;
 }
 
-/* find_bswap_1 invokes itself recursively with N and tries to perform
-   the operation given by the rhs of STMT on the result.  If the
-   operation could successfully be executed the function returns the
-   tree expression of the source operand and NULL otherwise.  */
+/* Check if STMT might be a byte swap or a nop from a memory source and returns
+   the answer. If so, REF is that memory source and the base of the memory area
+   accessed and the offset of the access from that base are recorded in N.  */
+
+bool
+find_bswap_or_nop_load (gimple stmt, tree ref, struct symbolic_number *n)
+{
+  /* Leaf node is an array or component ref. Memorize its base and
+     offset from base to compare to other such leaf node.  */
+  HOST_WIDE_INT bitsize, bitpos;
+  enum machine_mode mode;
+  int unsignedp, volatilep;
+
+  if (!gimple_assign_load_p (stmt) || gimple_has_volatile_ops (stmt))
+    return false;
+
+  n->base_addr = get_inner_reference (ref, &bitsize, &bitpos, &n->offset,
+				      &mode, &unsignedp, &volatilep, false);
+
+  if (TREE_CODE (n->base_addr) == MEM_REF)
+    {
+      offset_int bit_offset = 0;
+      tree off = TREE_OPERAND (n->base_addr, 1);
+
+      if (!integer_zerop (off))
+	{
+	  offset_int boff, coff = mem_ref_offset (n->base_addr);
+	  boff = wi::lshift (coff, LOG2_BITS_PER_UNIT);
+	  bit_offset += boff;
+	}
+
+      n->base_addr = TREE_OPERAND (n->base_addr, 0);
+
+      /* Avoid returning a negative bitpos as this may wreak havoc later.  */
+      if (wi::neg_p (bit_offset))
+	{
+	  offset_int mask = wi::mask <offset_int> (LOG2_BITS_PER_UNIT, false);
+	  offset_int tem = bit_offset.and_not (mask);
+	  /* TEM is the bitpos rounded to BITS_PER_UNIT towards -Inf.
+	     Subtract it to BIT_OFFSET and add it (scaled) to OFFSET.  */
+	  bit_offset -= tem;
+	  tem = wi::arshift (tem, LOG2_BITS_PER_UNIT);
+	  if (n->offset)
+	    n->offset = size_binop (PLUS_EXPR, n->offset,
+				    wide_int_to_tree (sizetype, tem));
+	  else
+	    n->offset = wide_int_to_tree (sizetype, tem);
+	}
+
+      bitpos += bit_offset.to_shwi ();
+    }
+
+  if (bitpos % BITS_PER_UNIT)
+    return false;
+  if (bitsize % BITS_PER_UNIT)
+    return false;
+
+  n->bytepos = bitpos / BITS_PER_UNIT;
+  n->alias_set = reference_alias_ptr_type (ref);
+  n->vuse = gimple_vuse (stmt);
+  return true;
+}
+
+/* find_bswap_or_nop_1 invokes itself recursively with N and tries to perform
+   the operation given by the rhs of STMT on the result.  If the operation
+   could successfully be executed the function returns the tree expression of
+   the source operand and NULL otherwise.  */
 
 static tree
-find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
+find_bswap_or_nop_1 (gimple stmt, struct symbolic_number *n, int limit)
 {
   enum tree_code code;
   tree rhs1, rhs2 = NULL;
@@ -1689,6 +1783,9 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
 
   rhs1 = gimple_assign_rhs1 (stmt);
 
+  if (find_bswap_or_nop_load (stmt, rhs1, n))
+    return rhs1;
+
   if (TREE_CODE (rhs1) != SSA_NAME)
     return NULL_TREE;
 
@@ -1715,11 +1812,11 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
 	  && code != CONVERT_EXPR)
 	return NULL_TREE;
 
-      source_expr1 = find_bswap_1 (rhs1_stmt, n, limit - 1);
+      source_expr1 = find_bswap_or_nop_1 (rhs1_stmt, n, limit - 1);
 
-      /* If find_bswap_1 returned NULL STMT is a leaf node and we have
-	 to initialize the symbolic number.  */
-      if (!source_expr1)
+      /* If find_bswap_or_nop_1 returned NULL, STMT is a leaf node and
+	 we have to initialize the symbolic number.  */
+      if (!source_expr1 || gimple_assign_load_p (rhs1_stmt))
 	{
 	  /* Set up the symbolic number N by setting each byte to a
 	     value between 1 and the byte size of rhs1.  The highest
@@ -1729,14 +1826,18 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
 	  if (n->size % BITS_PER_UNIT != 0)
 	    return NULL_TREE;
 	  n->size /= BITS_PER_UNIT;
-	  n->n = (sizeof (HOST_WIDEST_INT) < 8 ? 0 :
-		  (unsigned HOST_WIDEST_INT)0x08070605 << 32 | 0x04030201);
+	  n->range = n->size;
+	  n->n = CMPNOP;
 
 	  if (n->size < (int)sizeof (HOST_WIDEST_INT))
 	    n->n &= ((unsigned HOST_WIDEST_INT)1 <<
 		     (n->size * BITS_PER_UNIT)) - 1;
 
-	  source_expr1 = rhs1;
+	  if (!source_expr1)
+	    {
+	      n->base_addr = n->offset = n->alias_set = n->vuse = NULL_TREE;
+	      source_expr1 = rhs1;
+	    }
 	}
 
       switch (code)
@@ -1777,6 +1878,8 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
 		n->n &= ((unsigned HOST_WIDEST_INT)1 << type_size) - 1;
 	      }
 	    n->size = type_size / BITS_PER_UNIT;
+	    if (!n->base_addr)
+	      n->range = n->size;
 	  }
 	  break;
 	default:
@@ -1805,17 +1908,79 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
       switch (code)
 	{
 	case BIT_IOR_EXPR:
-	  source_expr1 = find_bswap_1 (rhs1_stmt, &n1, limit - 1);
+	  source_expr1 = find_bswap_or_nop_1 (rhs1_stmt, &n1, limit - 1);
 
 	  if (!source_expr1)
 	    return NULL_TREE;
 
-	  source_expr2 = find_bswap_1 (rhs2_stmt, &n2, limit - 1);
+	  source_expr2 = find_bswap_or_nop_1 (rhs2_stmt, &n2, limit - 1);
+
+	  if (n1.size != n2.size || !source_expr2)
+	    return NULL_TREE;
 
-	  if (source_expr1 != source_expr2
-	      || n1.size != n2.size)
+	  if (!n1.vuse != !n2.vuse ||
+	  (n1.vuse && !operand_equal_p (n1.vuse, n2.vuse, 0)))
 	    return NULL_TREE;
 
+	  if (source_expr1 != source_expr2)
+	    {
+	      HOST_WIDEST_INT inc, mask;
+	      unsigned i;
+	      HOST_WIDE_INT off_sub;
+	      struct symbolic_number *n_ptr;
+
+	      if (!n1.base_addr || !n2.base_addr
+		  || !operand_equal_p (n1.base_addr, n2.base_addr, 0))
+		return NULL_TREE;
+	      if (!n1.offset != !n2.offset ||
+	          (n1.offset && !operand_equal_p (n1.offset, n2.offset, 0)))
+		return NULL_TREE;
+
+	      /* We swap n1 with n2 to have n1 < n2.  */
+	      if (n2.bytepos < n1.bytepos)
+		{
+		  struct symbolic_number tmpn;
+
+		  tmpn = n2;
+		  n2 = n1;
+		  n1 = tmpn;
+		  source_expr1 = source_expr2;
+		}
+
+	      off_sub = n2.bytepos - n1.bytepos;
+
+	      /* Check that the range of memory covered < biggest int size.  */
+	      if (off_sub + n2.range > (int) sizeof (HOST_WIDEST_INT))
+	        return NULL_TREE;
+	      n->range = n2.range + off_sub;
+
+	      /* Reinterpret byte marks in symbolic number holding the value of
+		 bigger weight according to host endianness.  */
+	      inc = BYTES_BIG_ENDIAN ? off_sub + n2.range - n1.range : off_sub;
+	      mask = 0xFF;
+	      if (BYTES_BIG_ENDIAN)
+		n_ptr = &n1;
+	      else
+		n_ptr = &n2;
+	      for (i = 0; i < sizeof (HOST_WIDEST_INT); i++, inc <<= 8,
+		   mask <<= 8)
+		{
+		  if (n_ptr->n & mask)
+		    n_ptr->n += inc;
+		}
+	    }
+	  else
+	    n->range = n1.range;
+
+	  if (!n1.alias_set
+	      || alias_ptr_types_compatible_p (n1.alias_set, n2.alias_set))
+	    n->alias_set = n1.alias_set;
+	  else
+	    n->alias_set = ptr_type_node;
+	  n->vuse = n1.vuse;
+	  n->base_addr = n1.base_addr;
+	  n->offset = n1.offset;
+	  n->bytepos = n1.bytepos;
 	  n->size = n1.size;
 	  for (i = 0, mask = 0xff; i < n->size; i++, mask <<= BITS_PER_UNIT)
 	    {
@@ -1840,57 +2005,75 @@ find_bswap_1 (gimple stmt, struct symbolic_number *n, int limit)
   return NULL_TREE;
 }
 
-/* Check if STMT completes a bswap implementation consisting of ORs,
-   SHIFTs and ANDs.  Return the source tree expression on which the
-   byte swap is performed and NULL if no bswap was found.  */
+/* Check if STMT completes a bswap implementation or a read in a given
+   endianness consisting of ORs, SHIFTs and ANDs and sets *BSWAP
+   accordingly.  It also sets N to represent the kind of operations
+   performed: size of the resulting expression and whether it works on
+   a memory source, and if so alias-set and vuse.  At last, the
+   function returns the source tree expression.  */
 
 static tree
-find_bswap (gimple stmt)
+find_bswap_or_nop (gimple stmt, struct symbolic_number *n, bool *bswap)
 {
-/* The number which the find_bswap result should match in order to
-   have a full byte swap.  The number is shifted to the left according
-   to the size of the symbolic number before using it.  */
-  unsigned HOST_WIDEST_INT cmp =
-    sizeof (HOST_WIDEST_INT) < 8 ? 0 :
-    (unsigned HOST_WIDEST_INT)0x01020304 << 32 | 0x05060708;
-
-  struct symbolic_number n;
+/* The number which the find_bswap_or_nop_1 result should match in order
+   to have a full byte swap.  The number is shifted to the right
+   according to the size of the symbolic number before using it.  */
+  unsigned HOST_WIDEST_INT cmpxchg = CMPXCHG;
+  unsigned HOST_WIDEST_INT cmpnop = CMPNOP;
+
   tree source_expr;
   int limit;
 
   /* The last parameter determines the depth search limit.  It usually
-     correlates directly to the number of bytes to be touched.  We
-     increase that number by three  here in order to also
-     cover signed -> unsigned converions of the src operand as can be seen
+     correlates directly to the number n of bytes to be touched.  We
+     increase that number by log2(n) + 1 here in order to also
+     cover signed -> unsigned conversions of the src operand as can be seen
      in libgcc, and for initial shift/and operation of the src operand.  */
   limit = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (gimple_expr_type (stmt)));
   limit += 1 + (int) ceil_log2 ((unsigned HOST_WIDE_INT) limit);
-  source_expr =  find_bswap_1 (stmt, &n, limit);
+  source_expr =  find_bswap_or_nop_1 (stmt, n, limit);
 
   if (!source_expr)
     return NULL_TREE;
 
-  /* Zero out the extra bits of N and CMP.  */
-  if (n.size < (int)sizeof (HOST_WIDEST_INT))
+  /* Find real size of result (highest non zero byte).  */
+  if (n->base_addr)
     {
-      unsigned HOST_WIDEST_INT mask =
-	((unsigned HOST_WIDEST_INT)1 << (n.size * BITS_PER_UNIT)) - 1;
+      int rsize;
+      unsigned HOST_WIDEST_INT tmpn;
 
-      n.n &= mask;
-      cmp >>= (sizeof (HOST_WIDEST_INT) - n.size) * BITS_PER_UNIT;
+      for (tmpn = n->n, rsize = 0; tmpn; tmpn >>= BITS_PER_UNIT, rsize++);
+      n->range = rsize;
     }
 
-  /* A complete byte swap should make the symbolic number to start
-     with the largest digit in the highest order byte.  */
-  if (cmp != n.n)
+  /* Zero out the extra bits of N and CMP*.  */
+  if (n->range < (int)sizeof (HOST_WIDEST_INT))
+    {
+      unsigned HOST_WIDEST_INT mask;
+
+      mask = ((unsigned HOST_WIDEST_INT)1 << (n->range * BITS_PER_UNIT)) - 1;
+      cmpxchg >>= (sizeof (HOST_WIDEST_INT) - n->range) * BITS_PER_UNIT;
+      cmpnop &= mask;
+    }
+
+  /* A complete byte swap should make the symbolic number to start with
+     the largest digit in the highest order byte. Unchanged symbolic
+     number indicates a read with same endianness as host architecture.  */
+  if (n->n == cmpnop)
+    *bswap = false;
+  else if (n->n == cmpxchg)
+    *bswap = true;
+  else
     return NULL_TREE;
 
+  /* Useless bit manipulation performed by code.  */
+  if (!n->base_addr && n->n == cmpnop)
+    return NULL_TREE;
+
+  n->range *= BITS_PER_UNIT;
   return source_expr;
 }
 
-/* Find manual byte swap implementations and turn them into a bswap
-   builtin invokation.  */
-
 namespace {
 
 const pass_data pass_data_optimize_bswap =
@@ -1924,6 +2107,156 @@ public:
 
 }; // class pass_optimize_bswap
 
+/* Perform the bswap optimization: replace the statement STMT at GSI
+   with load type, VUSE and set-alias as described by N if a memory
+   source is involved (N->base_addr is non null), followed by the
+   builtin bswap invocation in FNDECL if BSWAP is true.  SRC gives
+   the source on which STMT is operating and N->range gives the
+   size of the expression involved for maintaining some statistics.  */
+
+static bool
+bswap_replace (gimple stmt, gimple_stmt_iterator *gsi, tree src, tree fndecl,
+	       tree bswap_type, tree load_type, struct symbolic_number *n,
+	       bool bswap)
+{
+  tree tmp, tgt;
+  gimple call;
+
+  tgt = gimple_assign_lhs (stmt);
+
+  /* Need to load the value from memory first.  */
+  if (n->base_addr)
+    {
+      tree addr_expr, addr_tmp, val_expr, val_tmp;
+      tree load_offset_ptr, aligned_load_type;
+      gimple addr_stmt, load_stmt;
+      unsigned align;
+
+      align = get_object_alignment (src);
+      if (bswap && SLOW_UNALIGNED_ACCESS (TYPE_MODE (load_type), align))
+	return false;
+
+      /*  Compute address to load from and cast according to the size
+	  of the load.  */
+      addr_expr = build_fold_addr_expr (unshare_expr (src));
+      if (is_gimple_min_invariant (addr_expr))
+	addr_tmp = addr_expr;
+      else
+	{
+	  addr_tmp = make_temp_ssa_name (TREE_TYPE (addr_expr), NULL,
+					 "load_src");
+	  addr_stmt = gimple_build_assign (addr_tmp, addr_expr);
+	  gsi_insert_before (gsi, addr_stmt, GSI_SAME_STMT);
+	}
+
+      /* Perform the load.  */
+      aligned_load_type = load_type;
+      if (align < TYPE_ALIGN (load_type))
+	aligned_load_type = build_aligned_type (load_type, align);
+      load_offset_ptr = build_int_cst (n->alias_set, 0);
+      val_expr = fold_build2 (MEM_REF, aligned_load_type, addr_tmp,
+			      load_offset_ptr);
+
+      if (!bswap)
+	{
+	  if (n->range == 16)
+	    nop_stats.found_16bit++;
+	  else if (n->range == 32)
+	    nop_stats.found_32bit++;
+	  else
+	    {
+	      gcc_assert (n->range == 64);
+	      nop_stats.found_64bit++;
+	    }
+
+	  /* Convert the result of load if necessary.  */
+	  if (!useless_type_conversion_p (TREE_TYPE (tgt), load_type))
+	    {
+	      val_tmp = make_temp_ssa_name (aligned_load_type, NULL,
+					    "load_dst");
+	      load_stmt = gimple_build_assign (val_tmp, val_expr);
+	      gimple_set_vuse (load_stmt, n->vuse);
+	      gsi_insert_before (gsi, load_stmt, GSI_SAME_STMT);
+	      gimple_assign_set_rhs_with_ops_1 (gsi, NOP_EXPR, val_tmp,
+						NULL_TREE, NULL_TREE);
+	    }
+	  else
+	    gimple_assign_set_rhs_with_ops_1 (gsi, MEM_REF, val_expr,
+					      NULL_TREE, NULL_TREE);
+	  update_stmt (gsi_stmt (*gsi));
+
+	  if (dump_file)
+	    {
+	      fprintf (dump_file,
+		       "%d bit load in host endianness found at: ",
+		       (int)n->range);
+	      print_gimple_stmt (dump_file, stmt, 0, 0);
+	    }
+	  return true;
+	}
+      else
+	{
+	  val_tmp = make_temp_ssa_name (aligned_load_type, NULL, "load_dst");
+	  load_stmt = gimple_build_assign (val_tmp, val_expr);
+	  gimple_set_vuse (load_stmt, n->vuse);
+	  gsi_insert_before (gsi, load_stmt, GSI_SAME_STMT);
+	}
+      src = val_tmp;
+    }
+
+  if (n->range == 16)
+    bswap_stats.found_16bit++;
+  else if (n->range == 32)
+    bswap_stats.found_32bit++;
+  else
+    {
+      gcc_assert (n->range == 64);
+      bswap_stats.found_64bit++;
+    }
+
+  tmp = src;
+
+  /* Convert the src expression if necessary.  */
+  if (!useless_type_conversion_p (TREE_TYPE (tmp), bswap_type))
+    {
+      gimple convert_stmt;
+      tmp = make_temp_ssa_name (bswap_type, NULL, "bswapsrc");
+      convert_stmt = gimple_build_assign_with_ops (NOP_EXPR, tmp, src, NULL);
+      gsi_insert_before (gsi, convert_stmt, GSI_SAME_STMT);
+    }
+
+  call = gimple_build_call (fndecl, 1, tmp);
+
+  tmp = tgt;
+
+  /* Convert the result if necessary.  */
+  if (!useless_type_conversion_p (TREE_TYPE (tgt), bswap_type))
+    {
+      gimple convert_stmt;
+      tmp = make_temp_ssa_name (bswap_type, NULL, "bswapdst");
+      convert_stmt = gimple_build_assign_with_ops (NOP_EXPR, tgt, tmp, NULL);
+      gsi_insert_after (gsi, convert_stmt, GSI_SAME_STMT);
+    }
+
+  gimple_call_set_lhs (call, tmp);
+
+  if (dump_file)
+    {
+      fprintf (dump_file, "%d bit bswap implementation found at: ",
+	       (int)n->range);
+      print_gimple_stmt (dump_file, stmt, 0, 0);
+    }
+
+  gsi_insert_after (gsi, call, GSI_SAME_STMT);
+  gsi_remove (gsi, true);
+  return true;
+}
+
+/* Find manual byte swap implementations as well as load in a given
+   endianness. Byte swaps are turned into a bswap builtin invokation
+   while endian loads are converted to bswap builtin invokation or
+   simple load according to the host endianness.  */
+
 unsigned int
 pass_optimize_bswap::execute (function *fun)
 {
@@ -1946,9 +2279,6 @@ pass_optimize_bswap::execute (function *fun)
 	       && (optab_handler (bswap_optab, DImode) != CODE_FOR_nothing
 		   || (bswap32_p && word_mode == SImode)));
 
-  if (!bswap16_p && !bswap32_p && !bswap64_p)
-    return 0;
-
   /* Determine the argument type of the builtins.  The code later on
      assumes that the return and argument type are the same.  */
   if (bswap16_p)
@@ -1969,6 +2299,7 @@ pass_optimize_bswap::execute (function *fun)
       bswap64_type = TREE_VALUE (TYPE_ARG_TYPES (TREE_TYPE (fndecl)));
     }
 
+  memset (&nop_stats, 0, sizeof (nop_stats));
   memset (&bswap_stats, 0, sizeof (bswap_stats));
 
   FOR_EACH_BB_FN (bb, fun)
@@ -1982,21 +2313,24 @@ pass_optimize_bswap::execute (function *fun)
       for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi); gsi_prev (&gsi))
         {
 	  gimple stmt = gsi_stmt (gsi);
-	  tree bswap_src, bswap_type;
-	  tree bswap_tmp;
-	  tree fndecl = NULL_TREE;
-	  int type_size;
-	  gimple call;
+	  tree fndecl = NULL_TREE, bswap_type = NULL_TREE;
+	  tree src, load_type;
+	  struct symbolic_number n;
+	  bool bswap;
 
 	  if (!is_gimple_assign (stmt)
 	      || gimple_assign_rhs_code (stmt) != BIT_IOR_EXPR)
 	    continue;
 
-	  type_size = TYPE_PRECISION (gimple_expr_type (stmt));
+	  src = find_bswap_or_nop (stmt, &n, &bswap);
+
+	  if (!src)
+	    continue;
 
-	  switch (type_size)
+	  switch (n.range)
 	    {
 	    case 16:
+	      load_type = uint16_type_node;
 	      if (bswap16_p)
 		{
 		  fndecl = builtin_decl_explicit (BUILT_IN_BSWAP16);
@@ -2004,6 +2338,7 @@ pass_optimize_bswap::execute (function *fun)
 		}
 	      break;
 	    case 32:
+	      load_type = uint32_type_node;
 	      if (bswap32_p)
 		{
 		  fndecl = builtin_decl_explicit (BUILT_IN_BSWAP32);
@@ -2011,6 +2346,7 @@ pass_optimize_bswap::execute (function *fun)
 		}
 	      break;
 	    case 64:
+	      load_type = uint64_type_node;
 	      if (bswap64_p)
 		{
 		  fndecl = builtin_decl_explicit (BUILT_IN_BSWAP64);
@@ -2021,62 +2357,21 @@ pass_optimize_bswap::execute (function *fun)
 	      continue;
 	    }
 
-	  if (!fndecl)
+	  if (bswap && !fndecl)
 	    continue;
 
-	  bswap_src = find_bswap (stmt);
-
-	  if (!bswap_src)
-	    continue;
-
-	  changed = true;
-	  if (type_size == 16)
-	    bswap_stats.found_16bit++;
-	  else if (type_size == 32)
-	    bswap_stats.found_32bit++;
-	  else
-	    bswap_stats.found_64bit++;
-
-	  bswap_tmp = bswap_src;
-
-	  /* Convert the src expression if necessary.  */
-	  if (!useless_type_conversion_p (TREE_TYPE (bswap_tmp), bswap_type))
-	    {
-	      gimple convert_stmt;
-	      bswap_tmp = make_temp_ssa_name (bswap_type, NULL, "bswapsrc");
-	      convert_stmt = gimple_build_assign_with_ops
-		  		(NOP_EXPR, bswap_tmp, bswap_src, NULL);
-	      gsi_insert_before (&gsi, convert_stmt, GSI_SAME_STMT);
-	    }
-
-	  call = gimple_build_call (fndecl, 1, bswap_tmp);
-
-	  bswap_tmp = gimple_assign_lhs (stmt);
-
-	  /* Convert the result if necessary.  */
-	  if (!useless_type_conversion_p (TREE_TYPE (bswap_tmp), bswap_type))
-	    {
-	      gimple convert_stmt;
-	      bswap_tmp = make_temp_ssa_name (bswap_type, NULL, "bswapdst");
-	      convert_stmt = gimple_build_assign_with_ops
-			(NOP_EXPR, gimple_assign_lhs (stmt), bswap_tmp, NULL);
-	      gsi_insert_after (&gsi, convert_stmt, GSI_SAME_STMT);
-	    }
-
-	  gimple_call_set_lhs (call, bswap_tmp);
-
-	  if (dump_file)
-	    {
-	      fprintf (dump_file, "%d bit bswap implementation found at: ",
-		       (int)type_size);
-	      print_gimple_stmt (dump_file, stmt, 0, 0);
-	    }
-
-	  gsi_insert_after (&gsi, call, GSI_SAME_STMT);
-	  gsi_remove (&gsi, true);
+	  if (bswap_replace (stmt, &gsi, src, fndecl, bswap_type, load_type,
+			     &n, bswap))
+	    changed = true;
 	}
     }
 
+  statistics_counter_event (fun, "16-bit nop implementations found",
+			    nop_stats.found_16bit);
+  statistics_counter_event (fun, "32-bit nop implementations found",
+			    nop_stats.found_32bit);
+  statistics_counter_event (fun, "64-bit nop implementations found",
+			    nop_stats.found_64bit);
   statistics_counter_event (fun, "16-bit bswap implementations found",
 			    bswap_stats.found_16bit);
   statistics_counter_event (fun, "32-bit bswap implementations found",