PATCH: Fix bugs in strcmp-sse4.S and strcmp-ssse3.S

["H.J. Lu" <hongjiu dot lu at intel dot com>]

Hi,

This patch adds a testcase for strncmp and fix bugs in strcmp-sse4.S and
strcmp-ssse3.S. These bugs do not affect the correctness of strcmp since
the bug is ifdefed by USE_AS_STRNCMP in strcmp-ssse3.S and in strcmp-sse4.S
we just compare the same bytes twice.


H.J.
---
2010-04-15  H.J. Lu  <hongjiu.lu@intel.com>

	* string/test-strncmp.c (check_result): New.
	(do_one_test): Use it.
	(check1): New.
	(test_main): Use it.

	* sysdeps/i386/i686/multiarch/strcmp-sse4.S (crosspage): Properly
	update source and destination.

	* sysdeps/i386/i686/multiarch/strcmp-ssse3.S (gobble_ashr_12):
	Properly check and update counter.

diff --git a/string/test-strncmp.c b/string/test-strncmp.c
index 5adf0eb..7cd90e3 100644
--- a/string/test-strncmp.c
+++ b/string/test-strncmp.c
@@ -51,8 +51,8 @@ stupid_strncmp (const char *s1, const char *s2, size_t n)
   return ret;
 }
 
-static void
-do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
+static int
+check_result (impl_t *impl, const char *s1, const char *s2, size_t n,
 	     int exp_result)
 {
   int result = CALL (impl, s1, s2, n);
@@ -63,9 +63,19 @@ do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
       error (0, 0, "Wrong result in function %s %d %d", impl->name,
 	     result, exp_result);
       ret = 1;
-      return;
+      return -1;
     }
 
+  return 0;
+}
+
+static void
+do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
+	     int exp_result)
+{
+  if (check_result (impl, s1, s2, n, exp_result) < 0)
+    return;
+
   if (HP_TIMING_AVAIL)
     {
       hp_timing_t start __attribute ((unused));
@@ -283,6 +293,25 @@ do_random_tests (void)
     }
 }
 
+static void
+check1 (void)
+{
+  char *s1 = (char *)(buf1 + 0xb2c);
+  char *s2 = (char *)(buf1 + 0xfd8);
+  size_t i;
+  int exp_result;
+
+  strcpy(s1, "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrs");
+  strcpy(s2, "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijkLMNOPQRSTUV");
+
+  for (i = 0; i < 80; i++)
+    {
+      exp_result = simple_strncmp (s1, s2, i);
+      FOR_EACH_IMPL (impl, 0)
+	 check_result (impl, s1, s2, i, exp_result);
+    }
+}
+
 int
 test_main (void)
 {
@@ -290,6 +319,8 @@ test_main (void)
 
   test_init ();
 
+  check1 ();
+
   printf ("%23s", "");
   FOR_EACH_IMPL (impl, 0)
     printf ("\t%s", impl->name);
diff --git a/sysdeps/i386/i686/multiarch/strcmp-sse4.S b/sysdeps/i386/i686/multiarch/strcmp-sse4.S
index 81d6ec6..0de0a11 100644
--- a/sysdeps/i386/i686/multiarch/strcmp-sse4.S
+++ b/sysdeps/i386/i686/multiarch/strcmp-sse4.S
@@ -223,8 +223,8 @@ L(crosspage):
 	inc	%edx
 	cmp	$15, %edx
 	jle	L(crosspage)
-	add	$16, %edi
-	add	$16, %esi
+	add	%edx, %edi
+	add	%edx, %esi
 	jmp	L(check_offset)
 
 	.p2align 4
diff --git a/sysdeps/i386/i686/multiarch/strcmp-ssse3.S b/sysdeps/i386/i686/multiarch/strcmp-ssse3.S
index 40994c0..a4de225 100644
--- a/sysdeps/i386/i686/multiarch/strcmp-ssse3.S
+++ b/sysdeps/i386/i686/multiarch/strcmp-ssse3.S
@@ -1484,17 +1484,18 @@ L(gobble_ashr_12):
 	sub	$0xffff, %esi
 	jnz	L(exit)
 
+#ifdef USE_AS_STRNCMP
+	cmp	$16, %ebp
+	lea	-16(%ebp), %ebp
+	jbe	L(more8byteseq)
+#endif
+
 	add	$16, %ecx
 	movdqa	%xmm4, %xmm3
 
 	add	$16, %edi
 	jg	L(nibble_ashr_12)
 
-#ifdef USE_AS_STRNCMP
-	cmp	$16, %ebp
-	lea	-16(%ebp), %ebp
-	jbe	L(more8byteseq)
-#endif
 	movdqa	(%eax, %ecx), %xmm1
 	movdqa	(%edx, %ecx), %xmm2
 	movdqa	%xmm2, %xmm4