tilegx: work around vector insn bug in gcc

Avoid an issue in gcc where some of the vector (aka SIMD) ops will
sometimes end up getting wrongly optimized out.  We use these
instructions in many of the string implementations.  If/when we
have an upstreamed fix for this problem in gcc we can conditionalize
the use of the extended assembly workaround in glibc.

9 files changed, 50 insertions, 25 deletions

diff --git a/sysdeps/tile/tilegx/memchr.c b/sysdeps/tile/tilegx/memchr.c
index 7da0f79da2..38c0da6737 100644
--- a/sysdeps/tile/tilegx/memchr.c
+++ b/sysdeps/tile/tilegx/memchr.c
@@ -58,7 +58,7 @@ __memchr (const void *s, int c, size_t n)
   /* Compute the address of the word containing the last byte. */
   last_word_ptr = (const uint64_t *) ((uintptr_t) last_byte_ptr & -8);
 
-  while ((bits = __insn_v1cmpeq (v, goal)) == 0)
+  while ((bits = v1cmpeq (v, goal)) == 0)
     {
       if (__builtin_expect (p == last_word_ptr, 0))
         {
diff --git a/sysdeps/tile/tilegx/rawmemchr.c b/sysdeps/tile/tilegx/rawmemchr.c
index 54b4a5c1b8..3f5044c83e 100644
--- a/sysdeps/tile/tilegx/rawmemchr.c
+++ b/sysdeps/tile/tilegx/rawmemchr.c
@@ -36,7 +36,7 @@ __rawmemchr (const void *s, int c)
   uint64_t v = (*p | before_mask) ^ (goal & before_mask);
 
   uint64_t bits;
-  while ((bits = __insn_v1cmpeq (v, goal)) == 0)
+  while ((bits = v1cmpeq (v, goal)) == 0)
     v = *++p;
 
   return ((char *) p) + (CFZ (bits) >> 3);
diff --git a/sysdeps/tile/tilegx/strchr.c b/sysdeps/tile/tilegx/strchr.c
index 36dfd31391..1a5eb5c927 100644
--- a/sysdeps/tile/tilegx/strchr.c
+++ b/sysdeps/tile/tilegx/strchr.c
@@ -38,16 +38,16 @@ strchr (const char *s, int c)
      match neither zero nor goal (we make sure the high bit of each byte
      is 1, and the low 7 bits are all the opposite of the goal byte).  */
   const uint64_t before_mask = MASK (s_int);
-  uint64_t v = (*p | before_mask) ^ (goal & __insn_v1shrui (before_mask, 1));
+  uint64_t v = (*p | before_mask) ^ (goal & v1shrui (before_mask, 1));
 
   uint64_t zero_matches, goal_matches;
   while (1)
     {
       /* Look for a terminating '\0'. */
-      zero_matches = __insn_v1cmpeqi (v, 0);
+      zero_matches = v1cmpeqi (v, 0);
 
       /* Look for the goal byte. */
-      goal_matches = __insn_v1cmpeq (v, goal);
+      goal_matches = v1cmpeq (v, goal);
 
       if (__builtin_expect ((zero_matches | goal_matches) != 0, 0))
         break;
diff --git a/sysdeps/tile/tilegx/strchrnul.c b/sysdeps/tile/tilegx/strchrnul.c
index e0f13b684e..e3024dda57 100644
--- a/sysdeps/tile/tilegx/strchrnul.c
+++ b/sysdeps/tile/tilegx/strchrnul.c
@@ -36,16 +36,16 @@ __strchrnul (const char *s, int c)
      match neither zero nor goal (we make sure the high bit of each byte
      is 1, and the low 7 bits are all the opposite of the goal byte).  */
   const uint64_t before_mask = MASK (s_int);
-  uint64_t v = (*p | before_mask) ^ (goal & __insn_v1shrui (before_mask, 1));
+  uint64_t v = (*p | before_mask) ^ (goal & v1shrui (before_mask, 1));
 
   uint64_t zero_matches, goal_matches;
   while (1)
     {
       /* Look for a terminating '\0'. */
-      zero_matches = __insn_v1cmpeqi (v, 0);
+      zero_matches = v1cmpeqi (v, 0);
 
       /* Look for the goal byte. */
-      goal_matches = __insn_v1cmpeq (v, goal);
+      goal_matches = v1cmpeq (v, goal);
 
       if (__builtin_expect ((zero_matches | goal_matches) != 0, 0))
         break;
diff --git a/sysdeps/tile/tilegx/string-endian.h b/sysdeps/tile/tilegx/string-endian.h
index fe9b073efb..6a3f882e60 100644
--- a/sysdeps/tile/tilegx/string-endian.h
+++ b/sysdeps/tile/tilegx/string-endian.h
@@ -56,3 +56,28 @@ static inline uint64_t copy_byte(uint8_t byte)
 {
   return __insn_shufflebytes(byte, 0, 0);
 }
+
+/* Implement the byte vector instructions using extended assembly.
+   The __insn_OP() builtins are buggy in current compiler versions.  */
+
+#define VECOP(OP)                                                       \
+  static inline uint64_t OP (uint64_t a, uint64_t b)                    \
+  {                                                                     \
+    uint64_t result;                                                    \
+    asm volatile (#OP " %0, %1, %2" : "=r"(result) : "r"(a), "r"(b));   \
+    return result;                                                      \
+  }                                                                     \
+                                                                        \
+  static inline uint64_t OP ## i (uint64_t a, uint64_t b)               \
+  {                                                                     \
+    uint64_t result;                                                    \
+    asm volatile (#OP "i %0, %1, %2" : "=r"(result) : "r"(a), "I"(b));  \
+    return result;                                                      \
+  }
+
+VECOP(v1cmpeq)
+VECOP(v1cmpltu)
+VECOP(v1cmpne)
+VECOP(v1add)
+VECOP(v1shru)
+VECOP(v1shl)
diff --git a/sysdeps/tile/tilegx/strlen.c b/sysdeps/tile/tilegx/strlen.c
index 5cd04acc59..cebdf22af5 100644
--- a/sysdeps/tile/tilegx/strlen.c
+++ b/sysdeps/tile/tilegx/strlen.c
@@ -31,7 +31,7 @@ strlen (const char *s)
   uint64_t v = *p | MASK (s_int);
 
   uint64_t bits;
-  while ((bits = __insn_v1cmpeqi (v, 0)) == 0)
+  while ((bits = v1cmpeqi (v, 0)) == 0)
     v = *++p;
 
   return ((const char *) p) + (CFZ (bits) >> 3) - s;
diff --git a/sysdeps/tile/tilegx/strnlen.c b/sysdeps/tile/tilegx/strnlen.c
index 5d73a14926..c3560d208d 100644
--- a/sysdeps/tile/tilegx/strnlen.c
+++ b/sysdeps/tile/tilegx/strnlen.c
@@ -37,7 +37,7 @@ __strnlen (const char *s, size_t maxlen)
   uint64_t v = *p | MASK (s_int);
 
   uint64_t bits;
-  while ((bits = __insn_v1cmpeqi (v, 0)) == 0)
+  while ((bits = v1cmpeqi (v, 0)) == 0)
     {
       if (bytes_read >= maxlen)
 	{
diff --git a/sysdeps/tile/tilegx/strrchr.c b/sysdeps/tile/tilegx/strrchr.c
index 5a9049e1b9..51a08b71bf 100644
--- a/sysdeps/tile/tilegx/strrchr.c
+++ b/sysdeps/tile/tilegx/strrchr.c
@@ -34,16 +34,16 @@ strrchr (const char *s, int c)
      match neither zero nor goal (we make sure the high bit of each byte
      is 1, and the low 7 bits are all the opposite of the goal byte).  */
   const uint64_t before_mask = MASK (s_int);
-  uint64_t v = (*p | before_mask) ^ (goal & __insn_v1shrui (before_mask, 1));
+  uint64_t v = (*p | before_mask) ^ (goal & v1shrui (before_mask, 1));
   const char *found = NULL;
   uint64_t zero_matches, goal_matches;
   while (1)
     {
       /* Look for a terminating '\0'. */
-      zero_matches = __insn_v1cmpeqi (v, 0);
+      zero_matches = v1cmpeqi (v, 0);
 
       /* Look for the goal byte. */
-      goal_matches = __insn_v1cmpeq (v, goal);
+      goal_matches = v1cmpeq (v, goal);
 
       /* If we found the goal, record the last offset. */
       if (__builtin_expect (goal_matches != 0, 0))
diff --git a/sysdeps/tile/tilegx/strstr.c b/sysdeps/tile/tilegx/strstr.c
index 548a92045d..f82936a3c2 100644
--- a/sysdeps/tile/tilegx/strstr.c
+++ b/sysdeps/tile/tilegx/strstr.c
@@ -57,10 +57,10 @@ static uint64_t
 vec_tolower (uint64_t cc)
 {
   /* For Uppercases letters, add 32 to convert to lower case.  */
-  uint64_t less_than_eq_Z = __insn_v1cmpltui (cc, 'Z' + 1);
-  uint64_t less_than_A =  __insn_v1cmpltui (cc, 'A');
-  uint64_t is_upper = __insn_v1cmpne (less_than_eq_Z, less_than_A);
-  return __insn_v1add (cc,__insn_v1shli (is_upper, 5));
+  uint64_t less_than_eq_Z = v1cmpltui (cc, 'Z' + 1);
+  uint64_t less_than_A =  v1cmpltui (cc, 'A');
+  uint64_t is_upper = v1cmpne (less_than_eq_Z, less_than_A);
+  return v1add (cc, v1shli (is_upper, 5));
 }
 
 /* There is no strcasechr() defined, but needed for 1 byte case
@@ -85,16 +85,16 @@ strcasechr (const char *s, int c)
      is 1, and the low 7 bits are all the opposite of the goal byte).  */
   const uint64_t before_mask = MASK (s_int);
   uint64_t v =
-    (vec_tolower (*p) | before_mask) ^ (goal & __insn_v1shrui (before_mask, 1));
+    (vec_tolower (*p) | before_mask) ^ (goal & v1shrui (before_mask, 1));
 
   uint64_t zero_matches, goal_matches;
   while (1)
     {
       /* Look for a terminating '\0'.  */
-      zero_matches = __insn_v1cmpeqi (v, 0);
+      zero_matches = v1cmpeqi (v, 0);
 
       /* Look for the goal byte.  */
-      goal_matches = __insn_v1cmpeq (v, goal);
+      goal_matches = v1cmpeq (v, goal);
 
       if (__builtin_expect ((zero_matches | goal_matches) != 0, 0))
         break;
@@ -146,14 +146,14 @@ STRSTR2 (const char *haystack_start, const char *needle)
      is 1, and the low 7 bits are all the opposite of the goal byte).  */
   const uint64_t before_mask = MASK (s_int);
   uint64_t v =
-    (vec_load (p) | before_mask) ^ (byte1 & __insn_v1shrui (before_mask, 1));
+    (vec_load (p) | before_mask) ^ (byte1 & v1shrui (before_mask, 1));
 
   uint64_t zero_matches, goal_matches;
   while (1)
     {
       /* Look for a terminating '\0'.  */
-      zero_matches = __insn_v1cmpeqi (v, 0);
-      uint64_t byte1_matches = __insn_v1cmpeq (v, byte1);
+      zero_matches = v1cmpeqi (v, 0);
+      uint64_t byte1_matches = v1cmpeq (v, byte1);
       if (__builtin_expect (zero_matches != 0, 0))
 	{
 	  /* This is the last vector.  Don't worry about matches
@@ -161,7 +161,7 @@ STRSTR2 (const char *haystack_start, const char *needle)
 	     back 1 byte to align it with the first byte, then and to
 	     check for both matching.  Each vector has a 1 in the LSB
 	     of the byte if there was match.  */
-	  uint64_t byte2_matches = __insn_v1cmpeq (v, byte2);
+	  uint64_t byte2_matches = v1cmpeq (v, byte2);
 	  goal_matches = byte1_matches & STRSHIFT (byte2_matches, 8);
 	  break;
 	}
@@ -175,7 +175,7 @@ STRSTR2 (const char *haystack_start, const char *needle)
 	    {
 	      /* 8-bytes starting 1 byte into v.  */
 	      v = __insn_dblalign (v, v2, (void*)1);
-	      uint64_t byte2_matches_shifted = __insn_v1cmpeq (v, byte2);
+	      uint64_t byte2_matches_shifted = v1cmpeq (v, byte2);
 	      goal_matches = byte1_matches & byte2_matches_shifted;
 	      if (__builtin_expect (goal_matches != 0, 0))
 		break;