Optimize x86/x86-64 disassembler tables.

3 files changed, 44 insertions, 11 deletions

diff --git a/libcpu/ChangeLog b/libcpu/ChangeLog
index 4b94d26a..a043e45f 100644
--- a/libcpu/ChangeLog
+++ b/libcpu/ChangeLog
@@ -1,5 +1,9 @@
 2009-01-01  Ulrich Drepper  <drepper@redhat.com>
 
+	* i386_parse.y (instrtable_out): Optimize match_data table by not
+	emitting 0xff masks for leading bytes.
+	* i386_disasm.c (i386_disasm): Adjust reader of match_data.
+
 	* i386_disasm.c (i386_disasm): Reset bufcnt when not matched.  We
 	don't expect snprintf to fail.
 
diff --git a/libcpu/i386_disasm.c b/libcpu/i386_disasm.c
index a656cdb4..c6bb0a58 100644
--- a/libcpu/i386_disasm.c
+++ b/libcpu/i386_disasm.c
@@ -1,5 +1,5 @@
 /* Disassembler for x86.
-   Copyright (C) 2007, 2008 Red Hat, Inc.
+   Copyright (C) 2007, 2008, 2009 Red Hat, Inc.
    This file is part of Red Hat elfutils.
    Written by Ulrich Drepper <drepper@redhat.com>, 2007.
 
@@ -376,37 +376,49 @@ i386_disasm (const uint8_t **startp, const uint8_t *end, GElf_Addr addr,
       while (curr < match_end)
 	{
 	  uint_fast8_t len = *curr++;
-	  const uint8_t *start = curr;
+	  uint_fast8_t clen = len >> 4;
+	  len &= 0xf;
+	  const uint8_t *next_curr = curr + clen + (len - clen) * 2;
 
 	  assert (len > 0);
-	  assert (curr + 2 * len <= match_end);
+	  assert (curr + clen + 2 * (len - clen) <= match_end);
 
 	  const uint8_t *codep = data;
 	  int correct_prefix = 0;
 	  int opoff = 0;
 
-	  if (data > begin && codep[-1] == curr[1] && curr[0] == 0xff)
+	  if (data > begin && codep[-1] == *curr && clen > 0)
 	    {
 	      /* We match a prefix byte.  This is exactly one byte and
 		 is matched exactly, without a mask.  */
 	      --len;
-	      start += 2;
+	      --clen;
 	      opoff = 8;
 
-	      curr += 2;
+	      ++curr;
 
 	      assert (last_prefix_bit != 0);
 	      correct_prefix = last_prefix_bit;
 	    }
 
 	  size_t avail = len;
+	  while (clen > 0)
+	    {
+	      if (*codep++ != *curr++)
+		goto not;
+	      --avail;
+	      --clen;
+	      if (codep == end && avail > 0)
+		goto do_ret;
+	    }
+
 	  while (avail > 0)
 	    {
 	      uint_fast8_t masked = *codep++ & *curr++;
 	      if (masked != *curr++)
 		{
 		not:
-		  curr = start + 2 * len;
+		  curr = next_curr;
 		  ++cnt;
 		  bufcnt = 0;
 		  goto next_match;
@@ -459,7 +471,7 @@ i386_disasm (const uint8_t **startp, const uint8_t *end, GElf_Addr addr,
 		 are not used uninitialized.  */
 	      asm (""
 		   : "=mr" (opoff), "=mr" (correct_prefix), "=mr" (codep),
-		     "=mr" (start), "=mr" (len));
+		     "=mr" (next_curr), "=mr" (len));
 	    }
 
 	  size_t prefix_size = 0;
diff --git a/libcpu/i386_parse.y b/libcpu/i386_parse.y
index b3a684d6..bea0e335 100644
--- a/libcpu/i386_parse.y
+++ b/libcpu/i386_parse.y
@@ -1,6 +1,6 @@
 %{
 /* Parser for i386 CPU description.
-   Copyright (C) 2004, 2005, 2007, 2008 Red Hat, Inc.
+   Copyright (C) 2004, 2005, 2007, 2008, 2009 Red Hat, Inc.
    Written by Ulrich Drepper <drepper@redhat.com>, 2004.
 
    Red Hat elfutils is free software; you can redistribute it and/or modify
@@ -37,6 +37,7 @@
 #include <math.h>
 #include <obstack.h>
 #include <search.h>
+#include <stdbool.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -1248,6 +1249,8 @@ instrtable_out (void)
       /* First count the number of bytes.  */
       size_t totalbits = 0;
       size_t zerobits = 0;
+      bool leading_p = true;
+      size_t leadingbits = 0;
       struct bitvalue *b = instr->bytes;
       while (b != NULL)
 	{
@@ -1255,6 +1258,8 @@ instrtable_out (void)
 	    {
 	      ++totalbits;
 	      zerobits = 0;
+	      if (leading_p)
+		++leadingbits;
 	    }
 	  else
 	    {
@@ -1264,13 +1269,15 @@ instrtable_out (void)
 		zerobits = 0;
 	      else
 		zerobits += b->field->bits;
+	      leading_p = false;
 	    }
 	  b = b->next;
 	}
       size_t nbytes = (totalbits - zerobits + 7) / 8;
       assert (nbytes > 0);
+      size_t leadingbytes = leadingbits / 8;
 
-      fprintf (outfile, "  %#zx,", nbytes);
+      fprintf (outfile, "  %#zx,", nbytes | (leadingbytes << 4));
 
       /* Now create the mask and byte values.  */
       uint8_t byte = 0;
@@ -1285,7 +1292,15 @@ instrtable_out (void)
 	      mask = (mask << 1) | 1;
 	      if (++nbits == 8)
 		{
-		  fprintf (outfile, " %#" PRIx8 ", %#" PRIx8 ",", mask, byte);
+		  if (leadingbytes > 0)
+		    {
+		      assert (mask == 0xff);
+		      fprintf (outfile, " %#" PRIx8 ",", byte);
+		      --leadingbytes;
+		    }
+		  else
+		    fprintf (outfile, " %#" PRIx8 ", %#" PRIx8 ",",
+			     mask, byte);
 		  byte = mask = nbits = 0;
 		  if (--nbytes == 0)
 		    break;
@@ -1293,6 +1308,8 @@ instrtable_out (void)
 	    }
 	  else
 	    {
+	      assert (leadingbytes == 0);
+
 	      unsigned long int remaining = b->field->bits;
 	      while (nbits + remaining > 8)
 		{