expr.c (MOVE_BY_PIECES_P): Define condition for deciding to use move_by_pieces.

	* expr.c (MOVE_BY_PIECES_P): Define condition for deciding to use
	move_by_pieces.
	(MOVE_MAX_PIECES): Define maximum number of bytes to move at once.
	(USE_LOAD_POST_INCREMENT, USE_LOAD_PRE_DECREMENT): Define defaults.
	(USE_STORE_POST_INCREMENT, USE_STORE_PRE_DECREMENT): Define defaults.
	(move_by_pieces): Use new macros.
	(emit_block_move): Use new macros.
	(clear_by_pieces): Use new macros.
	(clear_storage): Use new macros.
	(emit_push_insn): Use new macros.
	(expand_expr): Use new macros.
	* config/sh/sh.h (USE_LOAD_POST_INCREMENT, USE_LOAD_PRE_DECREMENT):
	Define.
	(USE_STORE_POST_INCREMENT, USE_STORE_PRE_DECREMENT): Define.
	(MOVE_BY_PIECES_P): Define based on alignment and TARGET_SMALLCODE.
	(MOVE_MAX_PIECES): move 8 bytes on SH4.
	* tm.texi(MOVE_BY_PIECES_P, MOVE_MAX_PIECES, USE_LOAD_POST_INCREMENT,
	USE_LOAD_PRE_DECREMENT, USE_STORE_POST_INCREMENT,
	USE_STORE_PRE_DECREMENT): Describe new macros.

From-SVN: r24801

4 files changed, 143 insertions, 27 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 9bc917b..69e53eb 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,25 @@
+Thu Jan 21 14:18:04 EST 1999  Andrew MacLeod  <amacleod@cygnus.com>
+
+	* expr.c (MOVE_BY_PIECES_P): Define condition for deciding to use
+	move_by_pieces.
+	(MOVE_MAX_PIECES): Define maximum number of bytes to move at once.
+	(USE_LOAD_POST_INCREMENT, USE_LOAD_PRE_DECREMENT): Define defaults.
+	(USE_STORE_POST_INCREMENT, USE_STORE_PRE_DECREMENT): Define defaults.
+	(move_by_pieces): Use new macros.
+	(emit_block_move): Use new macros.
+	(clear_by_pieces): Use new macros.
+	(clear_storage): Use new macros.
+	(emit_push_insn): Use new macros.
+	(expand_expr): Use new macros.
+	* config/sh/sh.h (USE_LOAD_POST_INCREMENT, USE_LOAD_PRE_DECREMENT): 
+	Define.
+	(USE_STORE_POST_INCREMENT, USE_STORE_PRE_DECREMENT): Define.
+	(MOVE_BY_PIECES_P): Define based on alignment and TARGET_SMALLCODE.
+	(MOVE_MAX_PIECES): move 8 bytes on SH4.
+	* tm.texi(MOVE_BY_PIECES_P, MOVE_MAX_PIECES, USE_LOAD_POST_INCREMENT,
+	USE_LOAD_PRE_DECREMENT, USE_STORE_POST_INCREMENT, 
+	USE_STORE_PRE_DECREMENT): Describe new macros.
+
 Thu Jan 21 14:13:31 1999  Vladimir N. Makarov  <vmakarov@cygnus.com>
 
         * varasm.c (output_constant_pool): Use floor_log2 instead of
diff --git a/gcc/config/sh/sh.h b/gcc/config/sh/sh.h
index e2bf8c7..260e9cb 100644
--- a/gcc/config/sh/sh.h
+++ b/gcc/config/sh/sh.h
@@ -1155,6 +1155,17 @@ extern struct rtx_def *sh_builtin_saveregs ();
 /*#define HAVE_POST_DECREMENT  1*/
 #define HAVE_PRE_DECREMENT   1
 
+#define USE_LOAD_POST_INCREMENT(mode)    ((mode == SImode || mode == DImode) \
+                                           ? 0 : 1)
+#define USE_LOAD_PRE_DECREMENT(mode)     0
+#define USE_STORE_POST_INCREMENT(mode)   0
+#define USE_STORE_PRE_DECREMENT(mode)    ((mode == SImode || mode == DImode) \
+                                           ? 0 : 1)
+
+#define MOVE_BY_PIECES_P(SIZE, ALIGN)  (move_by_pieces_ninsns (SIZE, ALIGN) \
+                                        < (TARGET_SMALLCODE ? 2 :           \
+                                           ((ALIGN >= 4) ? 16 : 2)))
+
 /* Macros to check register numbers against specific register classes.  */
 
 /* These assume that REGNO is a hard or pseudo reg number.
@@ -1509,6 +1520,10 @@ extern struct rtx_def *sh_builtin_saveregs ();
    in one reasonably fast instruction.  */
 #define MOVE_MAX 4
 
+/* Max number of bytes we want move_by_pieces to be able to copy
+   efficiently.  */
+#define MOVE_MAX_PIECES (TARGET_SH4 ? 8 : 4)
+
 /* Define if operations between registers always perform the operation
    on the full register even if a narrower mode is specified.  */
 #define WORD_REGISTER_OPERATIONS
@@ -2071,8 +2086,6 @@ extern int rtx_equal_function_value_matters;
 extern struct rtx_def *fpscr_rtx;
 extern struct rtx_def *get_fpscr_rtx ();
 
-
-#define MOVE_RATIO (TARGET_SMALLCODE ? 2 : 16)
 
 /* Instructions with unfilled delay slots take up an extra two bytes for
    the nop in the delay slot.  */
diff --git a/gcc/expr.c b/gcc/expr.c
index 1fd2afb..42e305d 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -210,6 +210,13 @@ static char direct_store[NUM_MACHINE_MODES];
 #endif
 #endif
 
+/* This macro is used to determine whether move_by_pieces should be called
+   to perform a structure copy. */
+#ifndef MOVE_BY_PIECES_P
+#define MOVE_BY_PIECES_P(SIZE, ALIGN) (move_by_pieces_ninsns        \
+                                       (SIZE, ALIGN) < MOVE_RATIO)
+#endif
+
 /* This array records the insn_code of insns to perform block moves.  */
 enum insn_code movstr_optab[NUM_MACHINE_MODES];
 
@@ -1383,6 +1390,38 @@ convert_modes (mode, oldmode, x, unsignedp)
   return temp;
 }
 
+
+/* This macro is used to determine what the largest unit size that
+   move_by_pieces can use is. */
+
+/* MOVE_MAX_PIECES is the number of bytes at a time which we can
+   move efficiently, as opposed to  MOVE_MAX which is the maximum
+   number of bhytes we can move with a single instruction. */
+
+#ifndef MOVE_MAX_PIECES
+#define MOVE_MAX_PIECES   MOVE_MAX
+#endif
+
+/* Some architectures do not have complete pre/post increment/decrement
+   instruction sets, or only move some modes efficiently. these macros
+   allow us to fine tune move_by_pieces for these targets. */
+
+#ifndef USE_LOAD_POST_INCREMENT
+#define USE_LOAD_POST_INCREMENT(MODE)   HAVE_POST_INCREMENT
+#endif
+
+#ifndef USE_LOAD_PRE_DECREMENT
+#define USE_LOAD_PRE_DECREMENT(MODE)    HAVE_PRE_DECREMENT
+#endif
+
+#ifndef USE_STORE_POST_INCREMENT
+#define USE_STORE_POST_INCREMENT(MODE)  HAVE_POST_INCREMENT
+#endif
+
+#ifndef USE_STORE_PRE_DECREMENT
+#define USE_STORE_PRE_DECREMENT(MODE)   HAVE_PRE_DECREMENT
+#endif
+
 /* 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
@@ -1396,7 +1435,9 @@ move_by_pieces (to, from, len, align)
 {
   struct move_by_pieces data;
   rtx to_addr = XEXP (to, 0), from_addr = XEXP (from, 0);
-  int max_size = MOVE_MAX + 1;
+  int max_size = MOVE_MAX_PIECES + 1;
+  enum machine_mode mode = VOIDmode, tmode;
+  enum insn_code icode;
 
   data.offset = 0;
   data.to_addr = to_addr;
@@ -1427,13 +1468,19 @@ move_by_pieces (to, from, len, align)
   if (!(data.autinc_from && data.autinc_to)
       && move_by_pieces_ninsns (len, align) > 2)
     {
-      if (HAVE_PRE_DECREMENT && data.reverse && ! data.autinc_from)
+      /* Find the mode of the largest move... */
+      for (tmode = GET_CLASS_NARROWEST_MODE (MODE_INT);
+	   tmode != VOIDmode; tmode = GET_MODE_WIDER_MODE (tmode))
+	if (GET_MODE_SIZE (tmode) < max_size)
+	  mode = tmode;
+
+      if (USE_LOAD_PRE_DECREMENT (mode) && data.reverse && ! data.autinc_from)
 	{
 	  data.from_addr = copy_addr_to_reg (plus_constant (from_addr, len));
 	  data.autinc_from = 1;
 	  data.explicit_inc_from = -1;
 	}
-      if (HAVE_POST_INCREMENT && ! data.autinc_from)
+      if (USE_LOAD_POST_INCREMENT (mode) && ! data.autinc_from)
 	{
 	  data.from_addr = copy_addr_to_reg (from_addr);
 	  data.autinc_from = 1;
@@ -1441,13 +1488,13 @@ move_by_pieces (to, from, len, align)
 	}
       if (!data.autinc_from && CONSTANT_P (from_addr))
 	data.from_addr = copy_addr_to_reg (from_addr);
-      if (HAVE_PRE_DECREMENT && data.reverse && ! data.autinc_to)
+      if (USE_STORE_PRE_DECREMENT (mode) && data.reverse && ! data.autinc_to)
 	{
 	  data.to_addr = copy_addr_to_reg (plus_constant (to_addr, len));
 	  data.autinc_to = 1;
 	  data.explicit_inc_to = -1;
 	}
-      if (HAVE_POST_INCREMENT && ! data.reverse && ! data.autinc_to)
+      if (USE_STORE_POST_INCREMENT (mode) && ! data.reverse && ! data.autinc_to)
 	{
 	  data.to_addr = copy_addr_to_reg (to_addr);
 	  data.autinc_to = 1;
@@ -1466,9 +1513,6 @@ move_by_pieces (to, from, len, align)
 
   while (max_size > 1)
     {
-      enum machine_mode mode = VOIDmode, tmode;
-      enum insn_code icode;
-
       for (tmode = GET_CLASS_NARROWEST_MODE (MODE_INT);
 	   tmode != VOIDmode; tmode = GET_MODE_WIDER_MODE (tmode))
 	if (GET_MODE_SIZE (tmode) < max_size)
@@ -1622,8 +1666,7 @@ emit_block_move (x, y, size, align)
   if (size == 0)
     abort ();
 
-  if (GET_CODE (size) == CONST_INT
-      && (move_by_pieces_ninsns (INTVAL (size), align) < MOVE_RATIO))
+  if (GET_CODE (size) == CONST_INT && MOVE_BY_PIECES_P (INTVAL (size), align))
     move_by_pieces (x, y, INTVAL (size), align);
   else
     {
@@ -2217,7 +2260,9 @@ clear_by_pieces (to, len, align)
 {
   struct clear_by_pieces data;
   rtx to_addr = XEXP (to, 0);
-  int max_size = MOVE_MAX + 1;
+  int max_size = MOVE_MAX_PIECES + 1;
+  enum machine_mode mode = VOIDmode, tmode;
+  enum insn_code icode;
 
   data.offset = 0;
   data.to_addr = to_addr;
@@ -2240,13 +2285,19 @@ clear_by_pieces (to, len, align)
   if (!data.autinc_to
       && move_by_pieces_ninsns (len, align) > 2)
     {
-      if (HAVE_PRE_DECREMENT && data.reverse && ! data.autinc_to)
+      /* Determine the main mode we'll be using */
+      for (tmode = GET_CLASS_NARROWEST_MODE (MODE_INT);
+	   tmode != VOIDmode; tmode = GET_MODE_WIDER_MODE (tmode))
+	if (GET_MODE_SIZE (tmode) < max_size)
+	  mode = tmode;
+
+      if (USE_STORE_PRE_DECREMENT (mode) && data.reverse && ! data.autinc_to)
 	{
 	  data.to_addr = copy_addr_to_reg (plus_constant (to_addr, len));
 	  data.autinc_to = 1;
 	  data.explicit_inc_to = -1;
 	}
-      if (HAVE_POST_INCREMENT && ! data.reverse && ! data.autinc_to)
+      if (USE_STORE_POST_INCREMENT (mode) && ! data.reverse && ! data.autinc_to)
 	{
 	  data.to_addr = copy_addr_to_reg (to_addr);
 	  data.autinc_to = 1;
@@ -2265,9 +2316,6 @@ clear_by_pieces (to, len, align)
 
   while (max_size > 1)
     {
-      enum machine_mode mode = VOIDmode, tmode;
-      enum insn_code icode;
-
       for (tmode = GET_CLASS_NARROWEST_MODE (MODE_INT);
 	   tmode != VOIDmode; tmode = GET_MODE_WIDER_MODE (tmode))
 	if (GET_MODE_SIZE (tmode) < max_size)
@@ -2351,7 +2399,7 @@ clear_storage (object, size, align)
       size = protect_from_queue (size, 0);
 
       if (GET_CODE (size) == CONST_INT
-	  && (move_by_pieces_ninsns (INTVAL (size), align) < MOVE_RATIO))
+	  && MOVE_BY_PIECES_P (INTVAL (size), align))
 	clear_by_pieces (object, INTVAL (size), align);
 
       else
@@ -2839,8 +2887,7 @@ emit_push_insn (x, mode, type, size, align, partial, reg, extra,
       if (args_addr == 0
 	  && GET_CODE (size) == CONST_INT
 	  && skip == 0
-	  && (move_by_pieces_ninsns ((unsigned) INTVAL (size) - used, align)
-	      < MOVE_RATIO)
+	  && (MOVE_BY_PIECES_P ((unsigned) INTVAL (size) - used, align)))
 	  /* Here we avoid the case of a structure whose weak alignment
 	     forces many pushes of a small amount of data,
 	     and such small pushes do rounding that causes trouble.  */
@@ -2938,8 +2985,7 @@ emit_push_insn (x, mode, type, size, align, partial, reg, extra,
 
 	  /* TEMP is the address of the block.  Copy the data there.  */
 	  if (GET_CODE (size) == CONST_INT
-	      && (move_by_pieces_ninsns ((unsigned) INTVAL (size), align)
-		  < MOVE_RATIO))
+	      && (MOVE_BY_PIECES_P ((unsigned) INTVAL (size), align)))
 	    {
 	      move_by_pieces (gen_rtx_MEM (BLKmode, temp), xinner,
 			      INTVAL (size), align);
@@ -6048,10 +6094,9 @@ expand_expr (exp, target, tmode, modifier)
 		     && ! (target != 0 && safe_from_p (target, exp, 1)))
 		    || TREE_ADDRESSABLE (exp)
 		    || (TREE_CODE (TYPE_SIZE (type)) == INTEGER_CST
-			&& (move_by_pieces_ninsns
-			    (TREE_INT_CST_LOW (TYPE_SIZE (type))/BITS_PER_UNIT,
-			     TYPE_ALIGN (type) / BITS_PER_UNIT)
-			    >= MOVE_RATIO)
+			&& (!MOVE_BY_PIECES_P 
+                             (TREE_INT_CST_LOW (TYPE_SIZE (type))/BITS_PER_UNIT,
+			     TYPE_ALIGN (type) / BITS_PER_UNIT))
 			&& ! mostly_zeros_p (exp))))
 	       || (modifier == EXPAND_INITIALIZER && TREE_CONSTANT (exp)))
 	{
diff --git a/gcc/tm.texi b/gcc/tm.texi
index cba1b66..2d7cd29 100644
--- a/gcc/tm.texi
+++ b/gcc/tm.texi
@@ -4754,6 +4754,42 @@ the corresponding number of memory-to-memory @emph{sequences}.
 
 If you don't define this, a reasonable default is used.
 
+@findex MOVE_BY_PIECES_P
+@item MOVE_BY_PIECES_P (@var{size}, @var{alignment})
+A C expression used to determine whether @code{move_by_pieces} will be used to
+copy a chunk of memory, or whether some other block move mechanism
+will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
+than @code{MOVE_RATIO}.
+
+@findex MOVE_MAX_PIECES
+@item MOVE_MAX_PIECES
+A C expression used by @code{move_by_pieces} to determine the largest unit
+a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
+
+@findex USE_LOAD_POST_INCREMENT
+@item USE_LOAD_POST_INCREMENT (@var{mode})
+A C expression used to determine whether a load postincrement is
+a good thing for @code{move_by_pieces} to use for a given mode. Defaults
+to the value of @code{HAVE_POST_INCREMENT}.
+
+@findex USE_LOAD_PRE_INCREMENT
+@item USE_LOAD_PRE_INCREMENT (@var{mode})
+A C expression used to determine whether a load preincrement is
+a good thing for @code{move_by_pieces} to use for a given mode. Defaults
+to the value of @code{HAVE_PRE_INCREMENT}.
+
+@findex USE_STORE_POST_INCREMENT
+@item USE_STORE_POST_INCREMENT (@var{mode})
+A C expression used to determine whether a store postincrement is
+a good thing for @code{move_by_pieces} to use for a given mode. Defaults
+to the value of @code{HAVE_POST_INCREMENT}.
+
+@findex USE_STORE_PRE_INCREMENT
+@item USE_STORE_PRE_INCREMENT (@var{mode})
+This macro is used to determine whether a store preincrement is
+a good thing for @code{move_by_pieces} to use for a given mode. Defaults
+to the value of @code{HAVE_PRE_INCREMENT}.
+
 @findex NO_FUNCTION_CSE
 @item NO_FUNCTION_CSE
 Define this macro if it is as good or better to call a constant