builtins.c (expand_builtin_memcpy): Remove endp argument and endp != 0 handling.

	* builtins.c (expand_builtin_memcpy): Remove endp argument and endp
	!= 0 handling.  Pass 0 to store_by_pieces.
	(expand_builtin_mempcpy): Add endp argument.  Don't call
	expand_builtin_memcpy, call store_by_pieces resp. move_by_pieces
	directly.  If ignoring result, only do expand_call.
	(expand_builtin_stpcpy): Likewise.  Call expand_builtin_mempcpy
	otherwise.
	(expand_builtin_strncpy, expand_builtin_memset): Adjust
	store_by_pices callers.
	(expand_builtin): Adjust expand_builtin_memcpy and
	expand_builtin_mempcpy callers.
	* expr.c (can_move_by_pieces): New function.
	(move_by_pieces): Add endp argument, return to resp. memory at end
	or one byte earlier depending on endp.
	(store_by_pieces): Likewise.
	(emit_block_move): Adjust call to move_by_pieces.
	(emit_push_insn): Adjust move_by_pieces caller.
	* expr.h (can_move_by_pieces): New prototype.
	(store_by_pieces): Adjust prototypes.
	* rtl.h (move_by_pieces): Adjust prototype.
	* config/mips/mips.c (expand_block_move): Adjust move_by_pieces
	caller.

	* gcc.c-torture/execute/builtins/string-4.c (main_test): Remove
	mempcpy test with post-increments.
	* gcc.c-torture/execute/string-opt-3.c: New test.
	* gcc.dg/string-opt-1.c: New test.

From-SVN: r67358

gcc/ChangeLog

+2003-06-03  Jakub Jelinek  <jakub@redhat.com>
+
+	* builtins.c (expand_builtin_memcpy): Remove endp argument and endp
+	!= 0 handling.  Pass 0 to store_by_pieces.
+	(expand_builtin_mempcpy): Add endp argument.  Don't call
+	expand_builtin_memcpy, call store_by_pieces resp. move_by_pieces
+	directly.  If ignoring result, only do expand_call.
+	(expand_builtin_stpcpy): Likewise.  Call expand_builtin_mempcpy
+	otherwise.
+	(expand_builtin_strncpy, expand_builtin_memset): Adjust
+	store_by_pices callers.
+	(expand_builtin): Adjust expand_builtin_memcpy and
+	expand_builtin_mempcpy callers.
+	* expr.c (can_move_by_pieces): New function.
+	(move_by_pieces): Add endp argument, return to resp. memory at end
+	or one byte earlier depending on endp.
+	(store_by_pieces): Likewise.
+	(emit_block_move): Adjust call to move_by_pieces.
+	(emit_push_insn): Adjust move_by_pieces caller.
+	* expr.h (can_move_by_pieces): New prototype.
+	(store_by_pieces): Adjust prototypes.
+	* rtl.h (move_by_pieces): Adjust prototype.
+	* config/mips/mips.c (expand_block_move): Adjust move_by_pieces
+	caller.
+
 2003-06-03  Ben Elliston  <bje@wasabisystems.com>
 
 	* doc/md.texi (Processor pipeline description): Improve wording.

gcc/config/mips/mips.c

@@ -3821,7 +3821,7 @@ expand_block_move (operands)
 
   else if (constp && bytes <= (unsigned)2 * MAX_MOVE_BYTES
 	   && align == (unsigned) UNITS_PER_WORD)
-    move_by_pieces (orig_dest, orig_src, bytes, align * BITS_PER_WORD);
+    move_by_pieces (orig_dest, orig_src, bytes, align * BITS_PER_WORD, 0);
 
   else if (constp && bytes <= (unsigned)2 * MAX_MOVE_BYTES)
     emit_insn (gen_movstrsi_internal (replace_equiv_address (orig_dest,

gcc/expr.c

@@ -1456,6 +1456,18 @@ convert_modes (mode, oldmode, x, unsignedp)
 
 #define STORE_MAX_PIECES  MIN (MOVE_MAX_PIECES, 2 * sizeof (HOST_WIDE_INT))
 
+/* Determine whether the LEN bytes can be moved by using several move
+   instructions.  Return nonzero if a call to move_by_pieces should
+   succeed.  */
+
+int
+can_move_by_pieces (len, align)
+     unsigned HOST_WIDE_INT len;
+     unsigned int align;
+{
+  return MOVE_BY_PIECES_P (len, align);
+}
+
 /* Generate several move instructions to copy LEN bytes from block FROM to
    block TO.  (These are MEM rtx's with BLKmode).  The caller must pass FROM
    and TO through protect_from_queue before calling.
@@ -1463,13 +1475,18 @@ convert_modes (mode, oldmode, x, unsignedp)
    If PUSH_ROUNDING is defined and TO is NULL, emit_single_push_insn is
    used to push FROM to the stack.
 
-   ALIGN is maximum stack alignment we can assume.  */
+   ALIGN is maximum stack alignment we can assume.
 
-void
-move_by_pieces (to, from, len, align)
+   If ENDP is 0 return to, if ENDP is 1 return memory at the end ala
+   mempcpy, and if ENDP is 2 return memory the end minus one byte ala
+   stpcpy.  */
+
+rtx
+move_by_pieces (to, from, len, align, endp)
      rtx to, from;
      unsigned HOST_WIDE_INT len;
      unsigned int align;
+     int endp;
 {
   struct move_by_pieces data;
   rtx to_addr, from_addr = XEXP (from, 0);
@@ -1583,6 +1600,36 @@ move_by_pieces (to, from, len, align)
   /* The code above should have handled everything.  */
   if (data.len > 0)
     abort ();
+
+  if (endp)
+    {
+      rtx to1;
+
+      if (data.reverse)
+	abort ();
+      if (data.autinc_to)
+	{
+	  if (endp == 2)
+	    {
+	      if (HAVE_POST_INCREMENT && data.explicit_inc_to > 0)
+		emit_insn (gen_add2_insn (data.to_addr, constm1_rtx));
+	      else
+		data.to_addr = copy_addr_to_reg (plus_constant (data.to_addr,
+								-1));
+	    }
+	  to1 = adjust_automodify_address (data.to, QImode, data.to_addr,
+					   data.offset);
+	}
+      else
+	{
+	  if (endp == 2)
+	    --data.offset;
+	  to1 = adjust_address (data.to, QImode, data.offset);
+	}
+      return to1;
+    }
+  else
+    return data.to;
 }
 
 /* Return number of insns required to move L bytes by pieces.
@@ -1760,7 +1807,7 @@ emit_block_move (x, y, size, method)
     }
 
   if (GET_CODE (size) == CONST_INT && MOVE_BY_PIECES_P (INTVAL (size), align))
-    move_by_pieces (x, y, INTVAL (size), align);
+    move_by_pieces (x, y, INTVAL (size), align, 0);
   else if (emit_block_move_via_movstr (x, y, size, align))
     ;
   else if (may_use_call)
@@ -2014,7 +2061,7 @@ init_block_move_fn (asmspec)
 {
   if (!block_move_fn)
     {
-      tree fn, args;
+      tree args, fn;
 
       if (TARGET_MEM_FUNCTIONS)
 	{
@@ -2738,15 +2785,19 @@ can_store_by_pieces (len, constfun, constfundata, align)
 /* Generate several move instructions to store LEN bytes generated by
    CONSTFUN to block TO.  (A MEM rtx with BLKmode).  CONSTFUNDATA is a
    pointer which will be passed as argument in every CONSTFUN call.
-   ALIGN is maximum alignment we can assume.  */
+   ALIGN is maximum alignment we can assume.
+   If ENDP is 0 return to, if ENDP is 1 return memory at the end ala
+   mempcpy, and if ENDP is 2 return memory the end minus one byte ala
+   stpcpy.  */
 
-void
-store_by_pieces (to, len, constfun, constfundata, align)
+rtx
+store_by_pieces (to, len, constfun, constfundata, align, endp)
      rtx to;
      unsigned HOST_WIDE_INT len;
      rtx (*constfun) PARAMS ((PTR, HOST_WIDE_INT, enum machine_mode));
      PTR constfundata;
      unsigned int align;
+     int endp;
 {
   struct store_by_pieces data;
 
@@ -2758,6 +2809,35 @@ store_by_pieces (to, len, constfun, constfundata, align)
   data.len = len;
   data.to = to;
   store_by_pieces_1 (&data, align);
+  if (endp)
+    {
+      rtx to1;
+
+      if (data.reverse)
+	abort ();
+      if (data.autinc_to)
+	{
+	  if (endp == 2)
+	    {
+	      if (HAVE_POST_INCREMENT && data.explicit_inc_to > 0)
+		emit_insn (gen_add2_insn (data.to_addr, constm1_rtx));
+	      else
+		data.to_addr = copy_addr_to_reg (plus_constant (data.to_addr,
+								-1));
+	    }
+	  to1 = adjust_automodify_address (data.to, QImode, data.to_addr,
+					   data.offset);
+	}
+      else
+	{
+	  if (endp == 2)
+	    --data.offset;
+	  to1 = adjust_address (data.to, QImode, data.offset);
+	}
+      return to1;
+    }
+  else
+    return data.to;
 }
 
 /* Generate several move instructions to clear LEN bytes of block TO.  (A MEM
@@ -3872,7 +3952,7 @@ emit_push_insn (x, mode, type, size, align, partial, reg, extra,
 	      && where_pad != none && where_pad != stack_direction)
 	    anti_adjust_stack (GEN_INT (extra));
 
-	  move_by_pieces (NULL, xinner, INTVAL (size) - used, align);
+	  move_by_pieces (NULL, xinner, INTVAL (size) - used, align, 0);
 	}
       else
 #endif /* PUSH_ROUNDING  */

gcc/expr.h

@@ -450,6 +450,11 @@ extern void use_group_regs PARAMS ((rtx *, rtx));
    If OBJECT has BLKmode, SIZE is its length in bytes.  */
 extern rtx clear_storage PARAMS ((rtx, rtx));
 
+/* Determine whether the LEN bytes can be moved by using several move
+   instructions.  Return nonzero if a call to move_by_pieces should
+   succeed.  */
+extern int can_move_by_pieces PARAMS ((unsigned HOST_WIDE_INT, unsigned int));
+
 /* Return nonzero if it is desirable to store LEN bytes generated by
    CONSTFUN with several move instructions by store_by_pieces
    function.  CONSTFUNDATA is a pointer which will be passed as argument
@@ -463,11 +468,12 @@ extern int can_store_by_pieces PARAMS ((unsigned HOST_WIDE_INT,
 /* Generate several move instructions to store LEN bytes generated by
    CONSTFUN to block TO.  (A MEM rtx with BLKmode).  CONSTFUNDATA is a
    pointer which will be passed as argument in every CONSTFUN call.
-   ALIGN is maximum alignment we can assume.  */
-extern void store_by_pieces PARAMS ((rtx, unsigned HOST_WIDE_INT,
-				     rtx (*) (PTR, HOST_WIDE_INT,
-					      enum machine_mode),
-				     PTR, unsigned int));
+   ALIGN is maximum alignment we can assume.
+   Returns TO + LEN.  */
+extern rtx store_by_pieces PARAMS ((rtx, unsigned HOST_WIDE_INT,
+				    rtx (*) (PTR, HOST_WIDE_INT,
+					     enum machine_mode),
+				    PTR, unsigned int, int));
 
 /* Emit insns to set X from Y.  */
 extern rtx emit_move_insn PARAMS ((rtx, rtx));

gcc/rtl.h

@@ -2141,9 +2141,9 @@ extern void emit_jump			PARAMS ((rtx));
 extern int preserve_subexpressions_p	PARAMS ((void));
 
 /* In expr.c */
-extern void move_by_pieces		PARAMS ((rtx, rtx,
+extern rtx move_by_pieces		PARAMS ((rtx, rtx,
 						 unsigned HOST_WIDE_INT,
-						 unsigned int));
+						 unsigned int, int));
 
 /* In flow.c */
 extern void recompute_reg_usage			PARAMS ((rtx, int));

gcc/testsuite/ChangeLog

+2003-06-03  Jakub Jelinek  <jakub@redhat.com>
+
+	* gcc.c-torture/execute/builtins/string-4.c (main_test): Remove
+	mempcpy test with post-increments.
+	* gcc.c-torture/execute/string-opt-3.c: New test.
+	* gcc.dg/string-opt-1.c: New test.
+
 2003-06-03  David Billinghurst (David.Billinghurst@riotinto.com)
 
 	PR fortran/10965

gcc/testsuite/gcc.c-torture/execute/builtins/string-4.c

@@ -23,7 +23,6 @@ void
 main_test (void)
 {
   int i;
-  const char *s;
 
   if (stpcpy (p, "abcde") != p + 5 || memcmp (p, "abcde", 6))
     abort ();
@@ -47,9 +46,8 @@ main_test (void)
   if (stpcpy ((i++, p + 20 + 1), "23") != (p + 20 + 1 + 2) || i != 9 || memcmp (p + 20, "q23\0u", 6))
     abort ();
 
-  s = s1; i = 3;
   memcpy (p + 25, "QRSTU", 6);
-  if (mempcpy (p + 25 + 1, s++, i++) != (p + 25 + 1 + 3) || i != 4 || s != s1 + 1 || memcmp (p + 25, "Q123U", 6))
+  if (mempcpy (p + 25 + 1, s1, 3) != (p + 25 + 1 + 3) || memcmp (p + 25, "Q123U", 6))
     abort ();
 
   if (stpcpy (stpcpy (p, "ABCD"), "EFG") != p + 7 || memcmp (p, "ABCDEFG", 8))

gcc/testsuite/gcc.c-torture/execute/string-opt-3.c

+/* Copyright (C) 2003  Free Software Foundation.
+
+   Ensure that builtin mempcpy and stpcpy perform correctly.
+
+   Written by Jakub Jelinek, 21/05/2003.  */
+
+extern void abort (void);
+typedef __SIZE_TYPE__ size_t;
+extern void *mempcpy (void *, const void *, size_t);
+extern int memcmp (const void *, const void *, size_t);
+extern char *stpcpy (char *, const char *);
+
+long buf1[64];
+char *buf2 = (char *) (buf1 + 32);
+long buf5[20];
+char buf7[20];
+
+int
+__attribute__((noinline))
+test (long *buf3, char *buf4, char *buf6, int n)
+{
+  int i = 0;
+
+  /* These should probably be handled by store_by_pieces on most arches.  */
+  if (mempcpy (buf1, "ABCDEFGHI", 9) != (char *) buf1 + 9
+      || memcmp (buf1, "ABCDEFGHI\0", 11))
+    abort ();
+
+  if (mempcpy (buf1, "abcdefghijklmnopq", 17) != (char *) buf1 + 17
+      || memcmp (buf1, "abcdefghijklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf3, "ABCDEF", 6) != (char *) buf1 + 6
+      || memcmp (buf1, "ABCDEFghijklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf3, "a", 1) != (char *) buf1 + 1
+      || memcmp (buf1, "aBCDEFghijklmnopq\0", 19))
+    abort ();
+
+  if (mempcpy ((char *) buf3 + 2, "bcd" + ++i, 2) != (char *) buf1 + 4
+      || memcmp (buf1, "aBcdEFghijklmnopq\0", 19)
+      || i != 1)
+    abort ();
+
+  /* These should probably be handled by move_by_pieces on most arches.  */
+  if (mempcpy ((char *) buf3 + 4, buf5, 6) != (char *) buf1 + 10
+      || memcmp (buf1, "aBcdRSTUVWklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy ((char *) buf1 + ++i + 8, (char *) buf5 + 1, 1)
+      != (char *) buf1 + 11
+      || memcmp (buf1, "aBcdRSTUVWSlmnopq\0", 19)
+      || i != 2)
+    abort ();
+
+  if (mempcpy ((char *) buf3 + 14, buf6, 2) != (char *) buf1 + 16
+      || memcmp (buf1, "aBcdRSTUVWSlmnrsq\0", 19))
+    abort ();
+
+  if (mempcpy (buf3, buf5, 8) != (char *) buf1 + 8
+      || memcmp (buf1, "RSTUVWXYVWSlmnrsq\0", 19))
+    abort ();
+
+  if (mempcpy (buf3, buf5, 17) != (char *) buf1 + 17
+      || memcmp (buf1, "RSTUVWXYZ01234567\0", 19))
+    abort ();
+
+  __builtin_memcpy (buf3, "aBcdEFghijklmnopq\0", 19);
+
+  /* These should be handled either by movstrendM or mempcpy
+     call.  */
+  if (mempcpy ((char *) buf3 + 4, buf5, n + 6) != (char *) buf1 + 10
+      || memcmp (buf1, "aBcdRSTUVWklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy ((char *) buf1 + ++i + 8, (char *) buf5 + 1, n + 1)
+      != (char *) buf1 + 12
+      || memcmp (buf1, "aBcdRSTUVWkSmnopq\0", 19)
+      || i != 3)
+    abort ();
+
+  if (mempcpy ((char *) buf3 + 14, buf6, n + 2) != (char *) buf1 + 16
+      || memcmp (buf1, "aBcdRSTUVWkSmnrsq\0", 19))
+    abort ();
+
+  i = 1;
+
+  /* These might be handled by store_by_pieces.  */
+  if (mempcpy (buf2, "ABCDEFGHI", 9) != buf2 + 9
+      || memcmp (buf2, "ABCDEFGHI\0", 11))
+    abort ();
+
+  if (mempcpy (buf2, "abcdefghijklmnopq", 17) != buf2 + 17
+      || memcmp (buf2, "abcdefghijklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf4, "ABCDEF", 6) != buf2 + 6
+      || memcmp (buf2, "ABCDEFghijklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf4, "a", 1) != buf2 + 1
+      || memcmp (buf2, "aBCDEFghijklmnopq\0", 19))
+    abort ();
+
+  if (mempcpy (buf4 + 2, "bcd" + i++, 2) != buf2 + 4
+      || memcmp (buf2, "aBcdEFghijklmnopq\0", 19)
+      || i != 2)
+    abort ();
+
+  /* These might be handled by move_by_pieces.  */
+  if (mempcpy (buf4 + 4, buf7, 6) != buf2 + 10
+      || memcmp (buf2, "aBcdRSTUVWklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf2 + i++ + 8, buf7 + 1, 1)
+      != buf2 + 11
+      || memcmp (buf2, "aBcdRSTUVWSlmnopq\0", 19)
+      || i != 3)
+    abort ();
+
+  if (mempcpy (buf4 + 14, buf6, 2) != buf2 + 16
+      || memcmp (buf2, "aBcdRSTUVWSlmnrsq\0", 19))
+    abort ();
+
+  __builtin_memcpy (buf4, "aBcdEFghijklmnopq\0", 19);
+
+  /* These should be handled either by movstrendM or mempcpy
+     call.  */
+  if (mempcpy (buf4 + 4, buf7, n + 6) != buf2 + 10
+      || memcmp (buf2, "aBcdRSTUVWklmnopq\0", 19))
+    abort ();
+
+  if (__builtin_mempcpy (buf2 + i++ + 8, buf7 + 1, n + 1)
+      != buf2 + 12
+      || memcmp (buf2, "aBcdRSTUVWkSmnopq\0", 19)
+      || i != 4)
+    abort ();
+
+  if (mempcpy (buf4 + 14, buf6, n + 2) != buf2 + 16
+      || memcmp (buf2, "aBcdRSTUVWkSmnrsq\0", 19))
+    abort ();
+
+  /* Now stpcpy tests.  */
+  if (stpcpy ((char *) buf3, "abcdefghijklmnop") != (char *) buf1 + 16
+      || memcmp (buf1, "abcdefghijklmnop", 17))
+    abort ();
+
+  if (__builtin_stpcpy ((char *) buf3, "ABCDEFG") != (char *) buf1 + 7
+      || memcmp (buf1, "ABCDEFG\0ijklmnop", 17))
+    abort ();
+
+  if (stpcpy ((char *) buf3 + i++, "x") != (char *) buf1 + 5
+      || memcmp (buf1, "ABCDx\0G\0ijklmnop", 17))
+    abort ();
+
+  return 0;
+}
+
+int
+main ()
+{
+  __builtin_memcpy (buf5, "RSTUVWXYZ0123456789", 20);
+  __builtin_memcpy (buf7, "RSTUVWXYZ0123456789", 20);
+  return test (buf1, buf2, "rstuvwxyz", 0);
+}

gcc/testsuite/gcc.dg/string-opt-1.c

+/* Ensure mempcpy is not "optimized" into memcpy followed by addition.  */
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+
+void *
+fn (char *x, char *y, int z)
+{
+  return __builtin_mempcpy (x, y, z);
+}
+
+/* { dg-final { scan-assembler-not "memcpy" } } */