CODEWITHSUNDEEP
c++clangllvminline-assemblyllvm-clang
KodeWarrior
KodeWarrior

Reputation: 3598

Clang error with inline assembly

  1 #ifndef ATOMIC_UTILS_H
  2 #define ATOMIC_UTILS_H
  3
  4 #include<cstddef>
  5
  6 class AtomicUtils
  7 {
  8     public:
  9
 10     /**
 11      * check if the value at addr is equal to oldval, if so replace it with newval
 12      * and return the oldval
 13      */
 14     static size_t compareAndExchange( volatile size_t* addr, size_t oldval, size_t newval )
 15     {
 16       size_t ret;
 17       __asm__ volatile( "lock cmpxchgl %2, %1\n\t"
 18                     :"=a"(ret), "+m"(*addr)
 19                     : "r"(newval), "0"(oldval)
 20                     : "memory" );
 21       return ret;
 22     }
 23 };
 24
 25 #endif

In file included from atomics_test.cpp:1:
./atomics.hpp:17:25: error: invalid operand for instruction
      __asm__ volatile( "lock cmpxchgl %2, %1\n\t"
                        ^
:1:16: note: instantiated into assembly here
        lock cmpxchgl %rsi, (%rdx)
                      ^~~~~

I'm totally clueless of the error. Could someone help plz ?? I'm using a 64 bit mac book pro with clang.

Upvotes: 3

Views: 1796

Answers (1)

Brian Walker
Brian Walker

Reputation: 8928

The error is that you are using cmpxchgl with 64-bit values (cmpxchgl wants 32-bit values). Try cmpxchgq for 64-bit.

Upvotes: 3

Related Questions