Fix gcc/sunc x86 AO_compare_double_and_swap_double missing side effect

(a cherry-pick of commit 43addfe from 'release-7_4')

Bind E/RDX to a dummy output since cmpxchg8/16b clobbers it.

* src/atomic_ops/sysdeps/gcc/x86.h [!__x86_64__]
(AO_compare_double_and_swap_double_full): New local variable (dummy);
specify "=d"(dummy) for asm code.
* src/atomic_ops/sysdeps/gcc/x86_64.h [AO_CMPXCHG16B_AVAILABLE]
(AO_compare_double_and_swap_double_full): Likewise.
* src/atomic_ops/sysdeps/sunc/x86_64.h [AO_CMPXCHG16B_AVAILABLE]
(AO_compare_double_and_swap_double_full): Likewise.
* src/atomic_ops/sysdeps/sunc/x86.h
(AO_compare_double_and_swap_double_full): Remove commented out
function.

4 files changed, 13 insertions, 44 deletions

diff --git a/src/atomic_ops/sysdeps/gcc/x86.h b/src/atomic_ops/sysdeps/gcc/x86.h
index 3348845..dfc4a26 100644
--- a/src/atomic_ops/sysdeps/gcc/x86.h
+++ b/src/atomic_ops/sysdeps/gcc/x86.h
@@ -138,6 +138,7 @@ AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
                                        AO_t old_val1, AO_t old_val2,
                                        AO_t new_val1, AO_t new_val2)
 {
+  AO_t dummy; /* an output for clobbered edx */
   char result;
 # ifdef __PIC__
     AO_t saved_ebx;
@@ -153,11 +154,12 @@ AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
 #   ifdef __OPTIMIZE__
       __asm__ __volatile__("mov %%ebx, %2\n\t" /* save ebx */
                            "lea %0, %%edi\n\t" /* in case addr is in ebx */
-                           "mov %7, %%ebx\n\t" /* load new_val1 */
+                           "mov %8, %%ebx\n\t" /* load new_val1 */
                            "lock; cmpxchg8b (%%edi)\n\t"
                            "mov %2, %%ebx\n\t" /* restore ebx */
                            "setz %1"
-                        : "=m" (*addr), "=a" (result), "=m" (saved_ebx)
+                        : "=m" (*addr), "=a" (result),
+                          "=m" (saved_ebx), "=d" (dummy)
                         : "m" (*addr), "d" (old_val2), "a" (old_val1),
                           "c" (new_val2), "m" (new_val1)
                         : "%edi", "memory");
@@ -169,13 +171,13 @@ AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
       __asm__ __volatile__("mov %%edi, %3\n\t" /* save edi */
                            "mov %%ebx, %2\n\t" /* save ebx */
                            "lea %0, %%edi\n\t" /* in case addr is in ebx */
-                           "mov %8, %%ebx\n\t" /* load new_val1 */
+                           "mov %9, %%ebx\n\t" /* load new_val1 */
                            "lock; cmpxchg8b (%%edi)\n\t"
                            "mov %2, %%ebx\n\t" /* restore ebx */
                            "mov %3, %%edi\n\t" /* restore edi */
                            "setz %1"
                         : "=m" (*addr), "=a" (result),
-                          "=m" (saved_ebx), "=m" (saved_edi)
+                          "=m" (saved_ebx), "=m" (saved_edi), "=d" (dummy)
                         : "m" (*addr), "d" (old_val2), "a" (old_val1),
                           "c" (new_val2), "m" (new_val1) : "memory");
 #   endif
@@ -184,7 +186,7 @@ AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
     /* faster) by using ebx as new_val1 (GCC would refuse to compile    */
     /* such code for PIC mode).                                         */
     __asm__ __volatile__("lock; cmpxchg8b %0; setz %1"
-                       : "=m" (*addr), "=a" (result)
+                       : "=m" (*addr), "=a" (result), "=d" (dummy)
                        : "m" (*addr), "d" (old_val2), "a" (old_val1),
                          "c" (new_val2), "b" (new_val1) : "memory");
 # endif
diff --git a/src/atomic_ops/sysdeps/gcc/x86_64.h b/src/atomic_ops/sysdeps/gcc/x86_64.h
index b2e920a..0eb8db4 100644
--- a/src/atomic_ops/sysdeps/gcc/x86_64.h
+++ b/src/atomic_ops/sysdeps/gcc/x86_64.h
@@ -142,9 +142,11 @@ AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
                                        AO_t old_val1, AO_t old_val2,
                                        AO_t new_val1, AO_t new_val2)
 {
+  AO_t dummy; /* an output for clobbered rdx */
   char result;
+
   __asm__ __volatile__("lock; cmpxchg16b %0; setz %1"
-                       : "=m"(*addr), "=a"(result)
+                       : "=m" (*addr), "=a" (result), "=d" (dummy)
                        : "m"(*addr), "d" (old_val2), "a" (old_val1),
                          "c" (new_val2), "b" (new_val1) : "memory");
   return (int) result;
diff --git a/src/atomic_ops/sysdeps/sunc/x86.h b/src/atomic_ops/sysdeps/sunc/x86.h
index fd8e132..4158c02 100644
--- a/src/atomic_ops/sysdeps/sunc/x86.h
+++ b/src/atomic_ops/sysdeps/sunc/x86.h
@@ -124,41 +124,4 @@ AO_compare_and_swap_full (volatile AO_t *addr, AO_t old, AO_t new_val)
 }
 #define AO_HAVE_compare_and_swap_full
 
-#if 0
-/* FIXME: not tested (and probably wrong). Besides,     */
-/* it tickles a bug in Sun C 5.10 (when optimizing).    */
-/* Returns nonzero if the comparison succeeded. */
-/* Really requires at least a Pentium.          */
-AO_INLINE int
-AO_compare_double_and_swap_double_full(volatile AO_double_t *addr,
-                                       AO_t old_val1, AO_t old_val2,
-                                       AO_t new_val1, AO_t new_val2)
-{
-  char result;
-#if __PIC__
-  /* If PIC is turned on, we can't use %ebx as it is reserved for the
-     GOT pointer.  We can save and restore %ebx because GCC won't be
-     using it for anything else (such as any of the m operands) */
-  __asm__ __volatile__("pushl %%ebx;"   /* save ebx used for PIC GOT ptr */
-                       "movl %6,%%ebx;" /* move new_val1 to %ebx */
-                       "lock; cmpxchg8b %0; setz %1;"
-                       "pop %%ebx;"     /* restore %ebx */
-                       : "=m"(*addr), "=a"(result)
-                       : "m"(*addr), "d" (old_val2), "a" (old_val1),
-                         "c" (new_val2), "m" (new_val1) : "memory");
-#else
-  /* We can't just do the same thing in non-PIC mode, because GCC
-   * might be using %ebx as the memory operand.  We could have ifdef'd
-   * in a clobber, but there's no point doing the push/pop if we don't
-   * have to. */
-  __asm__ __volatile__("lock; cmpxchg8b %0; setz %1;"
-                       : "=m"(*addr), "=a"(result)
-                       : /* "m"(*addr), */ "d" (old_val2), "a" (old_val1),
-                         "c" (new_val2), "b" (new_val1) : "memory");
-#endif
-  return (int) result;
-}
-#define AO_HAVE_compare_double_and_swap_double_full
-#endif
-
 #include "../ao_t_is_int.h"
diff --git a/src/atomic_ops/sysdeps/sunc/x86_64.h b/src/atomic_ops/sysdeps/sunc/x86_64.h
index 2350410..d92da1c 100644
--- a/src/atomic_ops/sysdeps/sunc/x86_64.h
+++ b/src/atomic_ops/sysdeps/sunc/x86_64.h
@@ -139,9 +139,11 @@ AO_compare_double_and_swap_double_full (volatile AO_double_t *addr,
                                         AO_t old_val1, AO_t old_val2,
                                         AO_t new_val1, AO_t new_val2)
 {
+  AO_t dummy;
   char result;
+
   __asm__ __volatile__ ("lock; cmpxchg16b %0; setz %1"
-                        : "+m" (*addr), "=a" (result)
+                        : "+m" (*addr), "=a" (result), "=d" (dummy)
                         : "d" (old_val2), "a" (old_val1),
                           "c" (new_val2), "b" (new_val1) : "memory");
   return (int) result;