re PR target/21329 (optimize i386 block copy)

2005-05-06  Denis Vlasenko  <vda@port.imtp.ilyichevsk.odessa.ua>
	    Jakub Jelinek  <jakub@redhat.com>

	PR target/21329
	* config/i386/i386.c (ix86_expand_movmem): Don't use rep; movsb
	for -Os if (movsl;)*(movsw;)?(movsb;)? sequence is shorter.
	Don't use rep; movs{l,q} if the repetition count is really small,
	instead use a sequence of movs{l,q} instructions.

Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r99330

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 846b95f..1db6106 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,12 @@
+2005-05-06  Denis Vlasenko  <vda@port.imtp.ilyichevsk.odessa.ua>
+           Jakub Jelinek  <jakub@redhat.com>
+
+       PR target/21329
+       * config/i386/i386.c (ix86_expand_movmem): Don't use rep; movsb
+       for -Os if (movsl;)*(movsw;)?(movsb;)? sequence is shorter.
+       Don't use rep; movs{l,q} if the repetition count is really small,
+       instead use a sequence of movs{l,q} instructions.
+
 2005-05-06  Jeff Law  <law@redhat.com>
 
        PR tree-optimization/21380

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 49d22fd..1e8326d 100644 (file)
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -11285,9 +11285,20 @@ ix86_expand_movmem (rtx dst, rtx src, rtx count_exp, rtx align_exp)
     src = replace_equiv_address_nv (src, srcreg);
 
   /* When optimizing for size emit simple rep ; movsb instruction for
-     counts not divisible by 4.  */
+     counts not divisible by 4, except when (movsl;)*(movsw;)?(movsb;)?
+     sequence is shorter than mov{b,l} $count, %{ecx,cl}; rep; movsb.
+     Sice of (movsl;)*(movsw;)?(movsb;)? sequence is
+     count / 4 + (count & 3), the other sequence is either 4 or 7 bytes,
+     but we don't know whether upper 24 (resp. 56) bits of %ecx will be
+     known to be zero or not.  The rep; movsb sequence causes higher
+     register preasure though, so take that into account.  */
 
-  if ((!optimize || optimize_size) && (count == 0 || (count & 0x03)))
+  if ((!optimize || optimize_size)
+      && (count == 0
+         || ((count & 0x03)
+             && (!optimize_size
+                 || count > 5 * 4
+                 || (count & 3) + count / 4 > 6))))
     {
       emit_insn (gen_cld ());
       countreg = ix86_zero_extend_to_Pmode (count_exp);
@@ -11313,19 +11324,36 @@ ix86_expand_movmem (rtx dst, rtx src, rtx count_exp, rtx align_exp)
       emit_insn (gen_cld ());
       if (count & ~(size - 1))
        {
-         countreg = copy_to_mode_reg (counter_mode,
-                                      GEN_INT ((count >> (size == 4 ? 2 : 3))
-                                               & (TARGET_64BIT ? -1 : 0x3fffffff)));
-         countreg = ix86_zero_extend_to_Pmode (countreg);
+         if ((TARGET_SINGLE_STRINGOP || optimize_size) && count < 5 * 4)
+           {
+             enum machine_mode movs_mode = size == 4 ? SImode : DImode;
+
+             while (offset < (count & ~(size - 1)))
+               {
+                 srcmem = adjust_automodify_address_nv (src, movs_mode,
+                                                        srcreg, offset);
+                 dstmem = adjust_automodify_address_nv (dst, movs_mode,
+                                                        destreg, offset);
+                 emit_insn (gen_strmov (destreg, dstmem, srcreg, srcmem));
+                 offset += size;
+               }
+           }
+         else
+           {
+             countreg = GEN_INT ((count >> (size == 4 ? 2 : 3))
+                                 & (TARGET_64BIT ? -1 : 0x3fffffff));
+             countreg = copy_to_mode_reg (counter_mode, countreg);
+             countreg = ix86_zero_extend_to_Pmode (countreg);
 
-         destexp = gen_rtx_ASHIFT (Pmode, countreg,
-                                   GEN_INT (size == 4 ? 2 : 3));
-         srcexp = gen_rtx_PLUS (Pmode, destexp, srcreg);
-         destexp = gen_rtx_PLUS (Pmode, destexp, destreg);
+             destexp = gen_rtx_ASHIFT (Pmode, countreg,
+                                       GEN_INT (size == 4 ? 2 : 3));
+             srcexp = gen_rtx_PLUS (Pmode, destexp, srcreg);
+             destexp = gen_rtx_PLUS (Pmode, destexp, destreg);
 
-         emit_insn (gen_rep_mov (destreg, dst, srcreg, src,
-                                 countreg, destexp, srcexp));
-         offset = count & ~(size - 1);
+             emit_insn (gen_rep_mov (destreg, dst, srcreg, src,
+                                     countreg, destexp, srcexp));
+             offset = count & ~(size - 1);
+           }
        }
       if (size == 8 && (count & 0x04))
        {