config.gcc (sh-*): Add sh-mem.o to extra_obj.

2013-10-25  Christian Bruel  <christian.bruel@st.com>

	* config.gcc (sh-*): Add sh-mem.o to extra_obj.
	* gcc/config/sh/t-sh (sh-mem.o): New rule.
	* gcc/config/sh/sh-mem.cc (expand_block_move): Moved here.
	(sh4_expand_cmpstr): New function.
	* gcc/config/sh/sh.c (force_into, expand_block_move): Move to sh-mem.cc
	* gcc/config/sh/sh-protos.h (sh4_expand_cmpstr): Declare.
	* gcc/config/sh/sh.md (cmpstrsi, cmpstr_t): New patterns.
	(rotlhi3_8): Rename.

From-SVN: r204013

7 files changed, 359 insertions, 157 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 5729910..49839c5 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,14 @@
+2013-10-25  Christian Bruel  <christian.bruel@st.com>
+
+	* config.gcc (sh-*): Add sh-mem.o to extra_obj.
+	* gcc/config/sh/t-sh (sh-mem.o): New rule.
+	* gcc/config/sh/sh-mem.cc (expand_block_move): Moved here.
+	(sh4_expand_cmpstr): New function.
+	* gcc/config/sh/sh.c (force_into, expand_block_move): Move to sh-mem.c
+	* gcc/config/sh/sh-protos.h (sh4_expand_cmpstr): Declare.
+	* gcc/config/sh/sh.md (cmpstrsi, cmpstr_t): New patterns.
+	(rotlhi3_8): Rename.
+
 2013-10-24  Jan-Benedict Glaw  <jbglaw@lug-owl.de>
 
 	* configure.ac (ZW_PROG_COMPILER_DEPENDENCIES): Use CXX instead of CC.
diff --git a/gcc/config.gcc b/gcc/config.gcc
index 022701d..a90ab25 100644
--- a/gcc/config.gcc
+++ b/gcc/config.gcc
@@ -465,7 +465,7 @@ sh[123456789lbe]*-*-* | sh-*-*)
 	cpu_type=sh
 	need_64bit_hwint=yes
 	extra_options="${extra_options} fused-madd.opt"
-	extra_objs="${extra_objs} sh_treg_combine.o"
+	extra_objs="${extra_objs} sh_treg_combine.o sh-mem.o"
 	;;
 v850*-*-*)
 	cpu_type=v850
diff --git a/gcc/config/sh/sh-mem.cc b/gcc/config/sh/sh-mem.cc
new file mode 100644
index 0000000..e6f0843
--- /dev/null
+++ b/gcc/config/sh/sh-mem.cc
@@ -0,0 +1,307 @@
+/* Helper routines for memory move and comparison insns.
+   Copyright (C) 2013 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 "tm.h"
+#include "machmode.h"
+#include "rtl.h"
+#include "tree.h"
+#include "expr.h"
+#include "tm_p.h"
+#include "basic-block.h"
+
+/* Like force_operand, but guarantees that VALUE ends up in TARGET.  */
+static void
+force_into (rtx value, rtx target)
+{
+  value = force_operand (value, target);
+  if (! rtx_equal_p (value, target))
+    emit_insn (gen_move_insn (target, value));
+}
+
+/* Emit code to perform a block move.  Choose the best method.
+
+   OPERANDS[0] is the destination.
+   OPERANDS[1] is the source.
+   OPERANDS[2] is the size.
+   OPERANDS[3] is the alignment safe to use.  */
+bool
+expand_block_move (rtx *operands)
+{
+  int align = INTVAL (operands[3]);
+  int constp = (CONST_INT_P (operands[2]));
+  int bytes = (constp ? INTVAL (operands[2]) : 0);
+
+  if (! constp)
+    return false;
+
+  /* If we could use mov.l to move words and dest is word-aligned, we
+     can use movua.l for loads and still generate a relatively short
+     and efficient sequence.  */
+  if (TARGET_SH4A_ARCH && align < 4
+      && MEM_ALIGN (operands[0]) >= 32
+      && can_move_by_pieces (bytes, 32))
+    {
+      rtx dest = copy_rtx (operands[0]);
+      rtx src = copy_rtx (operands[1]);
+      /* We could use different pseudos for each copied word, but
+	 since movua can only load into r0, it's kind of
+	 pointless.  */
+      rtx temp = gen_reg_rtx (SImode);
+      rtx src_addr = copy_addr_to_reg (XEXP (src, 0));
+      int copied = 0;
+
+      while (copied + 4 <= bytes)
+	{
+	  rtx to = adjust_address (dest, SImode, copied);
+	  rtx from = adjust_automodify_address (src, BLKmode,
+						src_addr, copied);
+
+	  set_mem_size (from, 4);
+	  emit_insn (gen_movua (temp, from));
+	  emit_move_insn (src_addr, plus_constant (Pmode, src_addr, 4));
+	  emit_move_insn (to, temp);
+	  copied += 4;
+	}
+
+      if (copied < bytes)
+	move_by_pieces (adjust_address (dest, BLKmode, copied),
+			adjust_automodify_address (src, BLKmode,
+						   src_addr, copied),
+			bytes - copied, align, 0);
+
+      return true;
+    }
+
+  /* If it isn't a constant number of bytes, or if it doesn't have 4 byte
+     alignment, or if it isn't a multiple of 4 bytes, then fail.  */
+  if (align < 4 || (bytes % 4 != 0))
+    return false;
+
+  if (TARGET_HARD_SH4)
+    {
+      if (bytes < 12)
+	return false;
+      else if (bytes == 12)
+	{
+	  rtx func_addr_rtx = gen_reg_rtx (Pmode);
+	  rtx r4 = gen_rtx_REG (SImode, 4);
+	  rtx r5 = gen_rtx_REG (SImode, 5);
+
+	  function_symbol (func_addr_rtx, "__movmemSI12_i4", SFUNC_STATIC);
+	  force_into (XEXP (operands[0], 0), r4);
+	  force_into (XEXP (operands[1], 0), r5);
+	  emit_insn (gen_block_move_real_i4 (func_addr_rtx));
+	  return true;
+	}
+      else if (! optimize_size)
+	{
+	  const char *entry_name;
+	  rtx func_addr_rtx = gen_reg_rtx (Pmode);
+	  int dwords;
+	  rtx r4 = gen_rtx_REG (SImode, 4);
+	  rtx r5 = gen_rtx_REG (SImode, 5);
+	  rtx r6 = gen_rtx_REG (SImode, 6);
+
+	  entry_name = (bytes & 4 ? "__movmem_i4_odd" : "__movmem_i4_even");
+	  function_symbol (func_addr_rtx, entry_name, SFUNC_STATIC);
+	  force_into (XEXP (operands[0], 0), r4);
+	  force_into (XEXP (operands[1], 0), r5);
+
+	  dwords = bytes >> 3;
+	  emit_insn (gen_move_insn (r6, GEN_INT (dwords - 1)));
+	  emit_insn (gen_block_lump_real_i4 (func_addr_rtx));
+	  return true;
+	}
+      else
+	return false;
+    }
+  if (bytes < 64)
+    {
+      char entry[30];
+      rtx func_addr_rtx = gen_reg_rtx (Pmode);
+      rtx r4 = gen_rtx_REG (SImode, 4);
+      rtx r5 = gen_rtx_REG (SImode, 5);
+
+      sprintf (entry, "__movmemSI%d", bytes);
+      function_symbol (func_addr_rtx, entry, SFUNC_STATIC);
+      force_into (XEXP (operands[0], 0), r4);
+      force_into (XEXP (operands[1], 0), r5);
+      emit_insn (gen_block_move_real (func_addr_rtx));
+      return true;
+    }
+
+  /* This is the same number of bytes as a memcpy call, but to a different
+     less common function name, so this will occasionally use more space.  */
+  if (! optimize_size)
+    {
+      rtx func_addr_rtx = gen_reg_rtx (Pmode);
+      int final_switch, while_loop;
+      rtx r4 = gen_rtx_REG (SImode, 4);
+      rtx r5 = gen_rtx_REG (SImode, 5);
+      rtx r6 = gen_rtx_REG (SImode, 6);
+
+      function_symbol (func_addr_rtx, "__movmem", SFUNC_STATIC);
+      force_into (XEXP (operands[0], 0), r4);
+      force_into (XEXP (operands[1], 0), r5);
+
+      /* r6 controls the size of the move.  16 is decremented from it
+	 for each 64 bytes moved.  Then the negative bit left over is used
+	 as an index into a list of move instructions.  e.g., a 72 byte move
+	 would be set up with size(r6) = 14, for one iteration through the
+	 big while loop, and a switch of -2 for the last part.  */
+
+      final_switch = 16 - ((bytes / 4) % 16);
+      while_loop = ((bytes / 4) / 16 - 1) * 16;
+      emit_insn (gen_move_insn (r6, GEN_INT (while_loop + final_switch)));
+      emit_insn (gen_block_lump_real (func_addr_rtx));
+      return true;
+    }
+
+  return false;
+}
+
+/* Emit code to perform a strcmp.
+
+   OPERANDS[0] is the destination.
+   OPERANDS[1] is the first string.
+   OPERANDS[2] is the second string.
+   OPERANDS[3] is the align.  */
+bool
+sh_expand_cmpstr (rtx *operands)
+{
+  rtx s1 = copy_rtx (operands[1]);
+  rtx s2 = copy_rtx (operands[2]);
+  rtx s1_addr = copy_addr_to_reg (XEXP (s1, 0));
+  rtx s2_addr = copy_addr_to_reg (XEXP (s2, 0));
+  rtx tmp0 = gen_reg_rtx (SImode);
+  rtx tmp1 = gen_reg_rtx (SImode);
+  rtx tmp2 = gen_reg_rtx (SImode);
+  rtx tmp3 = gen_reg_rtx (SImode);
+
+  rtx L_return = gen_label_rtx ();
+  rtx L_loop_byte = gen_label_rtx ();
+  rtx L_end_loop_byte = gen_label_rtx ();
+  rtx L_loop_long = gen_label_rtx ();
+  rtx L_end_loop_long = gen_label_rtx ();
+
+  rtx jump, addr1, addr2;
+  int prob_unlikely = REG_BR_PROB_BASE / 10;
+  int prob_likely = REG_BR_PROB_BASE / 4;
+
+  emit_insn (gen_iorsi3 (tmp1, s1_addr, s2_addr));
+  emit_move_insn (tmp0, GEN_INT (3));
+
+  emit_insn (gen_tstsi_t (tmp0, tmp1));
+
+  emit_move_insn (tmp0, const0_rtx);
+
+  jump = emit_jump_insn (gen_branch_false (L_loop_byte));
+  add_int_reg_note (jump, REG_BR_PROB, prob_likely);
+
+  addr1 = adjust_automodify_address (s1, SImode, s1_addr, 0);
+  addr2 = adjust_automodify_address (s2, SImode, s2_addr, 0);
+
+  /* tmp2 is aligned, OK to load.  */
+  emit_move_insn (tmp3, addr2);
+  emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 4));
+
+  /*start long loop.  */
+  emit_label (L_loop_long);
+
+  emit_move_insn (tmp2, tmp3);
+
+  /* tmp1 is aligned, OK to load.  */
+  emit_move_insn (tmp1, addr1);
+  emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 4));
+
+  /* Is there a 0 byte ?  */
+  emit_insn (gen_andsi3 (tmp3, tmp3, tmp1));
+
+  emit_insn (gen_cmpstr_t (tmp0, tmp3));
+  jump = emit_jump_insn (gen_branch_true (L_end_loop_long));
+  add_int_reg_note (jump, REG_BR_PROB, prob_unlikely);
+
+  emit_insn (gen_cmpeqsi_t (tmp1, tmp2));
+
+  /* tmp2 is aligned, OK to load.  */
+  emit_move_insn (tmp3, addr2);
+  emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 4));
+
+  jump = emit_jump_insn (gen_branch_true (L_loop_long));
+  add_int_reg_note (jump, REG_BR_PROB, prob_likely);
+  /* end loop.  */
+
+  /* Fallthu, check if one of the word is greater.  */
+  if (TARGET_LITTLE_ENDIAN)
+    {
+      rtx low_1 = gen_lowpart (HImode, tmp1);
+      rtx low_2 = gen_lowpart (HImode, tmp2);
+
+      emit_insn (gen_rotlhi3_8 (low_1, low_1));
+      emit_insn (gen_rotlhi3_8 (low_2, low_2));
+      emit_insn (gen_rotlsi3_16 (tmp1, tmp1));
+      emit_insn (gen_rotlsi3_16 (tmp2, tmp2));
+      emit_insn (gen_rotlhi3_8 (low_1, low_1));
+      emit_insn (gen_rotlhi3_8 (low_2, low_2));
+    }
+
+  jump = emit_jump_insn (gen_jump_compact (L_return));
+  emit_barrier_after (jump);
+
+  /* start byte loop.  */
+  addr1 = adjust_automodify_address (s1, QImode, s1_addr, 0);
+  addr2 = adjust_automodify_address (s2, QImode, s2_addr, 0);
+
+  emit_label (L_end_loop_long);
+
+  emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, -4));
+  emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, -4));
+
+  emit_label (L_loop_byte);
+
+  emit_insn (gen_extendqisi2 (tmp2, addr2));
+  emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 1));
+
+  emit_insn (gen_extendqisi2 (tmp1, addr1));
+  emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 1));
+
+  emit_insn (gen_cmpeqsi_t (tmp2, const0_rtx));
+  jump = emit_jump_insn (gen_branch_true (L_end_loop_byte));
+  add_int_reg_note (jump, REG_BR_PROB, prob_unlikely);
+
+  emit_insn (gen_cmpeqsi_t (tmp1, tmp2));
+  emit_jump_insn (gen_branch_true (L_loop_byte));
+  add_int_reg_note (jump, REG_BR_PROB, prob_likely);
+  /* end loop.  */
+
+  emit_label (L_end_loop_byte);
+
+  emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2)));
+  emit_insn (gen_zero_extendqisi2 (tmp1, gen_lowpart (QImode, tmp1)));
+
+  emit_label (L_return);
+
+  emit_insn (gen_subsi3 (operands[0], tmp1, tmp2));
+
+  return true;
+}
+
diff --git a/gcc/config/sh/sh-protos.h b/gcc/config/sh/sh-protos.h
index 4671c54..3484e4a 100644
--- a/gcc/config/sh/sh-protos.h
+++ b/gcc/config/sh/sh-protos.h
@@ -116,6 +116,7 @@ extern void emit_df_insn (rtx);
 extern void output_pic_addr_const (FILE *, rtx);
 extern bool expand_block_move (rtx *);
 extern void prepare_move_operands (rtx[], enum machine_mode mode);
+extern bool sh_expand_cmpstr (rtx *);
 extern enum rtx_code prepare_cbranch_operands (rtx *, enum machine_mode mode,
 					       enum rtx_code comparison);
 extern void expand_cbranchsi4 (rtx *operands, enum rtx_code comparison, int);
diff --git a/gcc/config/sh/sh.c b/gcc/config/sh/sh.c
index 31d1625..2e60763 100644
--- a/gcc/config/sh/sh.c
+++ b/gcc/config/sh/sh.c
@@ -174,7 +174,6 @@ static bool shmedia_space_reserved_for_target_registers;
 
 static void split_branches (rtx);
 static int branch_dest (rtx);
-static void force_into (rtx, rtx);
 static void print_slot (rtx);
 static rtx add_constant (rtx, enum machine_mode, rtx);
 static void dump_table (rtx, rtx);
@@ -1621,157 +1620,6 @@ sh_encode_section_info (tree decl, rtx rtl, int first)
     SYMBOL_REF_FLAGS (XEXP (rtl, 0)) |= SYMBOL_FLAG_FUNCVEC_FUNCTION;
 }
 
-/* Like force_operand, but guarantees that VALUE ends up in TARGET.  */
-static void
-force_into (rtx value, rtx target)
-{
-  value = force_operand (value, target);
-  if (! rtx_equal_p (value, target))
-    emit_insn (gen_move_insn (target, value));
-}
-
-/* Emit code to perform a block move.  Choose the best method.
-
-   OPERANDS[0] is the destination.
-   OPERANDS[1] is the source.
-   OPERANDS[2] is the size.
-   OPERANDS[3] is the alignment safe to use.  */
-bool
-expand_block_move (rtx *operands)
-{
-  int align = INTVAL (operands[3]);
-  int constp = (CONST_INT_P (operands[2]));
-  int bytes = (constp ? INTVAL (operands[2]) : 0);
-
-  if (! constp)
-    return false;
-
-  /* If we could use mov.l to move words and dest is word-aligned, we
-     can use movua.l for loads and still generate a relatively short
-     and efficient sequence.  */
-  if (TARGET_SH4A_ARCH && align < 4
-      && MEM_ALIGN (operands[0]) >= 32
-      && can_move_by_pieces (bytes, 32))
-    {
-      rtx dest = copy_rtx (operands[0]);
-      rtx src = copy_rtx (operands[1]);
-      /* We could use different pseudos for each copied word, but
-	 since movua can only load into r0, it's kind of
-	 pointless.  */
-      rtx temp = gen_reg_rtx (SImode);
-      rtx src_addr = copy_addr_to_reg (XEXP (src, 0));
-      int copied = 0;
-
-      while (copied + 4 <= bytes)
-	{
-	  rtx to = adjust_address (dest, SImode, copied);
-	  rtx from = adjust_automodify_address (src, BLKmode,
-						src_addr, copied);
-
-	  set_mem_size (from, 4);
-	  emit_insn (gen_movua (temp, from));
-	  emit_move_insn (src_addr, plus_constant (Pmode, src_addr, 4));
-	  emit_move_insn (to, temp);
-	  copied += 4;
-	}
-
-      if (copied < bytes)
-	move_by_pieces (adjust_address (dest, BLKmode, copied),
-			adjust_automodify_address (src, BLKmode,
-						   src_addr, copied),
-			bytes - copied, align, 0);
-
-      return true;
-    }
-
-  /* If it isn't a constant number of bytes, or if it doesn't have 4 byte
-     alignment, or if it isn't a multiple of 4 bytes, then fail.  */
-  if (align < 4 || (bytes % 4 != 0))
-    return false;
-
-  if (TARGET_HARD_SH4)
-    {
-      if (bytes < 12)
-	return false;
-      else if (bytes == 12)
-	{
-	  rtx func_addr_rtx = gen_reg_rtx (Pmode);
-	  rtx r4 = gen_rtx_REG (SImode, 4);
-	  rtx r5 = gen_rtx_REG (SImode, 5);
-
-	  function_symbol (func_addr_rtx, "__movmemSI12_i4", SFUNC_STATIC);
-	  force_into (XEXP (operands[0], 0), r4);
-	  force_into (XEXP (operands[1], 0), r5);
-	  emit_insn (gen_block_move_real_i4 (func_addr_rtx));
-	  return true;
-	}
-      else if (! optimize_size)
-	{
-	  const char *entry_name;
-	  rtx func_addr_rtx = gen_reg_rtx (Pmode);
-	  int dwords;
-	  rtx r4 = gen_rtx_REG (SImode, 4);
-	  rtx r5 = gen_rtx_REG (SImode, 5);
-	  rtx r6 = gen_rtx_REG (SImode, 6);
-
-	  entry_name = (bytes & 4 ? "__movmem_i4_odd" : "__movmem_i4_even");
-	  function_symbol (func_addr_rtx, entry_name, SFUNC_STATIC);
-	  force_into (XEXP (operands[0], 0), r4);
-	  force_into (XEXP (operands[1], 0), r5);
-
-	  dwords = bytes >> 3;
-	  emit_insn (gen_move_insn (r6, GEN_INT (dwords - 1)));
-	  emit_insn (gen_block_lump_real_i4 (func_addr_rtx));
-	  return true;
-	}
-      else
-	return false;
-    }
-  if (bytes < 64)
-    {
-      char entry[30];
-      rtx func_addr_rtx = gen_reg_rtx (Pmode);
-      rtx r4 = gen_rtx_REG (SImode, 4);
-      rtx r5 = gen_rtx_REG (SImode, 5);
-
-      sprintf (entry, "__movmemSI%d", bytes);
-      function_symbol (func_addr_rtx, entry, SFUNC_STATIC);
-      force_into (XEXP (operands[0], 0), r4);
-      force_into (XEXP (operands[1], 0), r5);
-      emit_insn (gen_block_move_real (func_addr_rtx));
-      return true;
-    }
-
-  /* This is the same number of bytes as a memcpy call, but to a different
-     less common function name, so this will occasionally use more space.  */
-  if (! optimize_size)
-    {
-      rtx func_addr_rtx = gen_reg_rtx (Pmode);
-      int final_switch, while_loop;
-      rtx r4 = gen_rtx_REG (SImode, 4);
-      rtx r5 = gen_rtx_REG (SImode, 5);
-      rtx r6 = gen_rtx_REG (SImode, 6);
-
-      function_symbol (func_addr_rtx, "__movmem", SFUNC_STATIC);
-      force_into (XEXP (operands[0], 0), r4);
-      force_into (XEXP (operands[1], 0), r5);
-
-      /* r6 controls the size of the move.  16 is decremented from it
-	 for each 64 bytes moved.  Then the negative bit left over is used
-	 as an index into a list of move instructions.  e.g., a 72 byte move
-	 would be set up with size(r6) = 14, for one iteration through the
-	 big while loop, and a switch of -2 for the last part.  */
-
-      final_switch = 16 - ((bytes / 4) % 16);
-      while_loop = ((bytes / 4) / 16 - 1) * 16;
-      emit_insn (gen_move_insn (r6, GEN_INT (while_loop + final_switch)));
-      emit_insn (gen_block_lump_real (func_addr_rtx));
-      return true;
-    }
-
-  return false;
-}
-
 /* Prepare operands for a move define_expand; specifically, one of the
    operands must be in a register.  */
 void
diff --git a/gcc/config/sh/sh.md b/gcc/config/sh/sh.md
index d8480cd..5642856 100644
--- a/gcc/config/sh/sh.md
+++ b/gcc/config/sh/sh.md
@@ -31,9 +31,6 @@
 ;; ??? The MAC.W and MAC.L instructions are not supported.  There is no
 ;; way to generate them.
 
-;; ??? The cmp/str instruction is not supported.  Perhaps it can be used
-;; for a str* inline function.
-
 ;; BSR is not generated by the compiler proper, but when relaxing, it
 ;; generates .uses pseudo-ops that allow linker relaxation to create
 ;; BSR.  This is actually implemented in bfd/{coff,elf32}-sh.c
@@ -4037,7 +4034,7 @@ label:
   DONE;
 })
 
-(define_insn "*rotlhi3_8"
+(define_insn "rotlhi3_8"
   [(set (match_operand:HI 0 "arith_reg_dest" "=r")
 	(rotate:HI (match_operand:HI 1 "arith_reg_operand" "r")
 		   (const_int 8)))]
@@ -11912,6 +11909,40 @@ label:
   "jsr	@%0%#"
   [(set_attr "type" "sfunc")
    (set_attr "needs_delay_slot" "yes")])
+
+;; byte compare pattern
+;; temp = a ^ b;
+;; !((temp & 0xF000) && (temp & 0x0F00) && (temp & 0x00F0) && (temp & 0x000F))
+(define_insn "cmpstr_t"
+  [(set (reg:SI T_REG)
+	(eq:SI (and:SI
+		(and:SI
+		 (and:SI
+		  (zero_extract:SI (xor:SI (match_operand:SI 0 "arith_reg_operand" "r")
+					   (match_operand:SI 1 "arith_reg_operand" "r"))
+				   (const_int 8) (const_int 0))
+		  (zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
+				   (const_int 8) (const_int 8)))
+		  (zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
+				   (const_int 8) (const_int 16)))
+		(zero_extract:SI (xor:SI (match_dup 0) (match_dup 1))
+				 (const_int 8) (const_int 24))) (const_int 0)))]
+  "TARGET_SH1"
+  "cmp/str	%0,%1"
+  [(set_attr "type" "mt_group")])
+
+(define_expand "cmpstrsi"
+  [(set (match_operand:SI 0 "register_operand")
+	(compare:SI (match_operand:BLK 1 "memory_operand")
+		    (match_operand:BLK 2 "memory_operand")))
+   (use (match_operand 3 "immediate_operand"))]
+  "TARGET_SH1"
+{
+   if (! optimize_insn_for_size_p () && sh_expand_cmpstr (operands))
+      DONE;
+   else FAIL;
+})
+
 
 ;; -------------------------------------------------------------------------
 ;; Floating point instructions.
diff --git a/gcc/config/sh/t-sh b/gcc/config/sh/t-sh
index 081ba3c..710bc87 100644
--- a/gcc/config/sh/t-sh
+++ b/gcc/config/sh/t-sh
@@ -16,6 +16,10 @@
 # along with GCC; see the file COPYING3.  If not see
 # <http://www.gnu.org/licenses/>.
 
+sh-mem.o: $(srcdir)/config/sh/sh-mem.cc \
+  $(CONFIG_H) $(SYSTEM_H) $(TREE_H) $(TM_H) $(TM_P_H)
+	$(COMPILER) -c $(ALL_COMPILERFLAGS) $(ALL_CPPFLAGS) $(INCLUDES) $<
+
 sh-c.o: $(srcdir)/config/sh/sh-c.c \
   $(CONFIG_H) $(SYSTEM_H) $(TREE_H) $(TM_H) $(TM_P_H) coretypes.h
 	$(COMPILER) -c $(ALL_COMPILERFLAGS) $(ALL_CPPFLAGS) $(INCLUDES) \