tree-optimization/103345: Improved load merging.

This patch implements PR tree-optimization/103345 to merge adjacent
loads when combined with addition or bitwise xor.  The current code
in gimple-ssa-store-merging.c's find_bswap_or_nop alreay handles ior,
so that all that's required is to treat PLUS_EXPR and BIT_XOR_EXPR in
the same way at BIT_IOR_EXPR.  Many thanks to Andrew Pinski for
pointing out that this also resolves PR target/98953.

2021-11-22  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog
	PR tree-optimization/98953
	PR tree-optimization/103345
	* gimple-ssa-store-merging.c (find_bswap_or_nop_1): Handle
	BIT_XOR_EXPR and PLUS_EXPR the same as BIT_IOR_EXPR.
	(pass_optimize_bswap::execute): Likewise.

gcc/testsuite/ChangeLog
	PR tree-optimization/98953
	PR tree-optimization/103345
	* gcc.dg/tree-ssa/pr98953.c: New test case.
	* gcc.dg/tree-ssa/pr103345.c: New test case.

gcc/gimple-ssa-store-merging.c

@@ -742,10 +742,7 @@ find_bswap_or_nop_1 (gimple *stmt, struct symbolic_number *n, int limit)
       struct symbolic_number n1, n2;
       gimple *source_stmt, *source_stmt2;
 
-      if (code != BIT_IOR_EXPR)
-	return NULL;
-
-      if (TREE_CODE (rhs2) != SSA_NAME)
+      if (!rhs2 || TREE_CODE (rhs2) != SSA_NAME)
 	return NULL;
 
       rhs2_stmt = SSA_NAME_DEF_STMT (rhs2);
@@ -753,6 +750,8 @@ find_bswap_or_nop_1 (gimple *stmt, struct symbolic_number *n, int limit)
       switch (code)
 	{
 	case BIT_IOR_EXPR:
+	case BIT_XOR_EXPR:
+	case PLUS_EXPR:
 	  source_stmt1 = find_bswap_or_nop_1 (rhs1_stmt, &n1, limit - 1);
 
 	  if (!source_stmt1)
@@ -1495,6 +1494,8 @@ pass_optimize_bswap::execute (function *fun)
 		continue;
 	      /* Fall through.  */
 	    case BIT_IOR_EXPR:
+	    case BIT_XOR_EXPR:
+	    case PLUS_EXPR:
 	      break;
 	    case CONSTRUCTOR:
 	      {

gcc/testsuite/gcc.dg/tree-ssa/pr103345.c

@@ -0,0 +1,53 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-bswap-details" } */
+
+typedef unsigned int uint32_t;
+typedef unsigned char uint8_t;
+
+uint32_t load_le_32_or(const uint8_t *ptr)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return ((uint32_t)ptr[0]) |
+         ((uint32_t)ptr[1] << 8) |
+         ((uint32_t)ptr[2] << 16) |
+         ((uint32_t)ptr[3] << 24);
+#else
+  return ((uint32_t)ptr[3]) |
+         ((uint32_t)ptr[2] << 8) |
+         ((uint32_t)ptr[1] << 16) |
+         ((uint32_t)ptr[0] << 24);
+#endif
+}
+
+uint32_t load_le_32_add(const uint8_t *ptr)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return ((uint32_t)ptr[0]) +
+         ((uint32_t)ptr[1] << 8) +
+         ((uint32_t)ptr[2] << 16) +
+         ((uint32_t)ptr[3] << 24);
+#else
+  return ((uint32_t)ptr[3]) +
+         ((uint32_t)ptr[2] << 8) +
+         ((uint32_t)ptr[1] << 16) +
+         ((uint32_t)ptr[0] << 24);
+#endif
+}
+
+uint32_t load_le_32_xor(const uint8_t *ptr)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  return ((uint32_t)ptr[0]) ^
+         ((uint32_t)ptr[1] << 8) ^
+         ((uint32_t)ptr[2] << 16) ^
+         ((uint32_t)ptr[3] << 24);
+#else
+  return ((uint32_t)ptr[0]) ^
+         ((uint32_t)ptr[1] << 8) ^
+         ((uint32_t)ptr[2] << 16) ^
+         ((uint32_t)ptr[3] << 24);
+#endif
+}
+
+/* { dg-final { scan-tree-dump-times "32 bit load in target endianness found" 3 "bswap" } } */
+

gcc/testsuite/gcc.dg/tree-ssa/pr98953.c

@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-bswap-details" } */
+
+int foo(unsigned char *ptr)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    return ptr[0] + (ptr[1] << 8);
+#else
+    return ptr[1] + (ptr[0] << 8);
+#endif
+}
+
+/* { dg-final { scan-tree-dump "16 bit load in target endianness found" "bswap" } } */
+