Fix gcc/sunc x86 AO_compare_double_and_swap_double missing side effect

(fix of commits 1dc2fb3, 929954a)

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

* src/atomic_ops/sysdeps/gcc/x86.h [!(AO_GCC_ATOMIC_TEST_AND_SET
&& !AO_SKIPATOMIC_double_compare_and_swap_ANY)]
(AO_compare_double_and_swap_double_full): Declare dummy local variable;
specify "=d"(dummy) for asm code.
* src/atomic_ops/sysdeps/sunc/x86.h [__i386 && !AO_NO_CMPXCHG8B ||
!__i386 && AO_CMPXCHG16B_AVAILABLE]
(AO_compare_double_and_swap_double_full): Likewise.

2 files changed, 15 insertions, 8 deletions

diff --git a/src/atomic_ops/sysdeps/gcc/x86.h b/src/atomic_ops/sysdeps/gcc/x86.h
index f68a8b0..57a1fdc 100644
--- a/src/atomic_ops/sysdeps/gcc/x86.h
+++ b/src/atomic_ops/sysdeps/gcc/x86.h
@@ -442,6 +442,7 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
                                          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;
@@ -457,11 +458,12 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
 #     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");
@@ -473,13 +475,13 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
         __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
@@ -488,7 +490,7 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
       /* 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");
@@ -552,9 +554,11 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
                                          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");
diff --git a/src/atomic_ops/sysdeps/sunc/x86.h b/src/atomic_ops/sysdeps/sunc/x86.h
index 88c6a78..dfac1e5 100644
--- a/src/atomic_ops/sysdeps/sunc/x86.h
+++ b/src/atomic_ops/sysdeps/sunc/x86.h
@@ -176,10 +176,11 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
                                            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;
 
       __asm__ __volatile__ ("lock; cmpxchg8b %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");
@@ -215,9 +216,11 @@ AO_fetch_compare_and_swap_full(volatile AO_t *addr, AO_t old_val,
                                             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");