sh-mem.cc (sh_expand_cmpnstr): Unroll small sizes and optimized non constant lengths.

2014-01-13  Christian Bruel  <christian.bruel@st.com>

	* config/sh/sh-mem.cc (sh_expand_cmpnstr): Unroll small sizes and
	  optimized non constant lengths.

From-SVN: r206575

gcc/ChangeLog

+2014-01-13  Christian Bruel  <christian.bruel@st.com>
+
+	* config/sh/sh-mem.cc (sh_expand_cmpnstr): Unroll small sizes and
+	  optimized non constant lengths.
+
 2014-01-13  Jakub Jelinek  <jakub@redhat.com>
 
 	PR libgomp/59194

gcc/config/sh/sh-mem.cc

@@ -324,7 +324,6 @@ sh_expand_cmpnstr (rtx *operands)
   rtx addr2 = operands[2];
   rtx s1_addr = copy_addr_to_reg (XEXP (addr1, 0));
   rtx s2_addr = copy_addr_to_reg (XEXP (addr2, 0));
-  rtx tmp0 = gen_reg_rtx (SImode);
   rtx tmp1 = gen_reg_rtx (SImode);
   rtx tmp2 = gen_reg_rtx (SImode);
 
@@ -334,22 +333,27 @@ sh_expand_cmpnstr (rtx *operands)
   rtx L_end_loop_byte = gen_label_rtx ();
 
   rtx len = force_reg (SImode, operands[3]);
-  int constp = (CONST_INT_P (operands[3]));
-  int bytes = (constp ? INTVAL (operands[3]) : 0);
-  int witers = bytes / 4;
+  int constp = CONST_INT_P (operands[3]);
 
-  /* We could still loop on a register count. Not found very
-     convincing to optimize yet.  */
-  if (! constp)
-    return false;
-
-  if (witers > 1)
+  /* Loop on a register count. */
+  if (constp)
     {
-      rtx L_loop_long = gen_label_rtx ();
-      rtx L_end_loop_long = gen_label_rtx ();
+      rtx tmp0 = gen_reg_rtx (SImode);
       rtx tmp3 = gen_reg_rtx (SImode);
       rtx lenw = gen_reg_rtx (SImode);
+
+      rtx L_loop_long = gen_label_rtx ();
+      rtx L_end_loop_long = gen_label_rtx ();
+      rtx L_small = gen_label_rtx ();
+
       int align = INTVAL (operands[4]);
+      int bytes = INTVAL (operands[3]);
+      int witers = bytes / 4;
+
+      if (witers > 1)
+        {
+          addr1 = adjust_automodify_address (addr1, SImode, s1_addr, 0);
+          addr2 = adjust_automodify_address (addr2, SImode, s2_addr, 0);
 
           emit_move_insn (tmp0, const0_rtx);
 
@@ -361,9 +365,6 @@ sh_expand_cmpnstr (rtx *operands)
               add_int_reg_note (jump, REG_BR_PROB, prob_likely);
             }
 
-      addr1 = adjust_automodify_address (addr1, SImode, s1_addr, 0);
-      addr2 = adjust_automodify_address (addr2, SImode, s2_addr, 0);
-
           /* word count. Do we have iterations ? */
           emit_insn (gen_lshrsi3 (lenw, len, GEN_INT (2)));
 
@@ -372,11 +373,13 @@ sh_expand_cmpnstr (rtx *operands)
 
           /* tmp2 is aligned, OK to load.  */
           emit_move_insn (tmp2, addr2);
-      emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 4));
+          emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr,
+                                                  GET_MODE_SIZE (SImode)));
 
           /* tmp1 is aligned, OK to load.  */
           emit_move_insn (tmp1, addr1);
-      emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 4));
+          emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr,
+                                                  GET_MODE_SIZE (SImode)));
 
           /* Is there a 0 byte ?  */
           emit_insn (gen_andsi3 (tmp3, tmp2, tmp1));
@@ -396,6 +399,7 @@ sh_expand_cmpnstr (rtx *operands)
               emit_insn (gen_addsi3 (lenw, lenw, GEN_INT (-1)));
               emit_insn (gen_tstsi_t (lenw, lenw));
             }
+
           jump = emit_jump_insn (gen_branch_false (L_loop_long));
           add_int_reg_note (jump, REG_BR_PROB, prob_likely);
 
@@ -409,23 +413,52 @@ sh_expand_cmpnstr (rtx *operands)
           else
             {
               /* Remaining bytes to read.   */
-	  emit_move_insn (len, GEN_INT (bytes % 4));
-	  jump = emit_jump_insn (gen_jump_compact (L_loop_byte));
+              jump = emit_jump_insn (gen_jump_compact (L_small));
               emit_barrier_after (jump);
             }
 
           emit_label (L_end_loop_long);
 
-      /* Remaining bytes to read.   */
-      emit_move_insn (len, GEN_INT (4));
-
           /* Found last word.  Restart it byte per byte. */
-      emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, -4));
-      emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, -4));
+          bytes =  4;
+          emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr,
+                                                  -GET_MODE_SIZE (SImode)));
+          emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr,
+                                                  -GET_MODE_SIZE (SImode)));
         }
 
-  addr1 = adjust_address (addr1, QImode, 0);
-  addr2 = adjust_address (addr2, QImode, 0);
+      emit_label (L_small);
+
+      gcc_assert (bytes <= 7);
+
+      addr1 = adjust_automodify_address (addr1, QImode, s1_addr, 0);
+      addr2 = adjust_automodify_address (addr2, QImode, s2_addr, 0);
+
+      while (bytes--)
+        {
+          emit_insn (gen_extendqisi2 (tmp1, addr1));
+          emit_insn (gen_extendqisi2 (tmp2, addr2));
+
+          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));
+          if (flag_delayed_branch)
+            emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2)));
+          jump = emit_jump_insn (gen_branch_false (L_end_loop_byte));
+          add_int_reg_note (jump, REG_BR_PROB, prob_unlikely);
+
+          addr1 = adjust_address (addr1, QImode, GET_MODE_SIZE (QImode));
+          addr2 = adjust_address (addr2, QImode, GET_MODE_SIZE (QImode));
+        }
+
+      jump = emit_jump_insn (gen_jump_compact( L_end_loop_byte));
+      emit_barrier_after (jump);
+    }
+
+  addr1 = adjust_automodify_address (addr1, QImode, s1_addr, 0);
+  addr2 = adjust_automodify_address (addr2, QImode, s2_addr, 0);
 
   emit_label (L_loop_byte);
 

gcc/testsuite/ChangeLog

+2014-01-13  Christian Bruel  <christian.bruel@st.com>
+
+	* gcc.target/sh/cmpstrn.c: New case.
+
 2014-01-13  Jakub Jelinek  <jakub@redhat.com>
 
 	* gcc.dg/vect/vect-simd-clone-10.c: Add dg-do run.

gcc/testsuite/gcc.target/sh/cmpstrn.c

@@ -6,16 +6,23 @@
 /* { dg-final { scan-assembler-not "jmp" } } */
 /* { dg-final { scan-assembler-times "cmp/str" 1 } } */
 
-/* Test that the cmp/str loop is optimized out.  */
-test01(const char *s1, const char *s2, int n)
+/* Test that cmp/str is not used for small lengths.  */
+test01(const char *s1)
 {
   return __builtin_strncmp (s1, "abcde", 3);
 }
 
 /* Test that the cmp/str loop is used.  */
-test02(const char *s1, const char *s2, int n)
+test02(const char *s1)
 {
   return __builtin_strncmp (s1, "abcdefghi", 8);
 }
 
+/* Test that no call is generated  */
+test03(const char *s1, int n)
+{
+  return __builtin_strncmp (s1, "abcde", n);
+}
+
+