Re: GCC in-line assembly and the removal of -mcx16

[Andrew Haley <aph at redhat dot com>]

On 30/05/17 15:52, Alexander Monakov wrote:
> On Tue, 30 May 2017, Jonathan Wakely wrote:
>>> I just built GCC from trunk, and this gives me:
>>>
>>> mustang-b0:~ $ /scratch/gcc/trunk/install/bin/gcc zzz.c
>>> /tmp/ccmc3Y73.o: In function `main':
>>> zzz.c:(.text+0x20): undefined reference to `__sync_val_compare_and_swap_16'
>>> collect2: error: ld returned 1 exit status
>>>
>>> I can't figure out where __sync_val_compare_and_swap_16 should be
>>> defined.  It's not in libatomic or libc or libgcc.
> 
> libgcc implements __sync..._N calls in terms of __sync_val_compare_and_swap_N
> built-in when it's available, but there's no way libgcc could conjure CAS "out
> of thin air".

But why is it not then in libgcc?  It's not hard to write it in a
way that will run on all AArch64 systems.

int __sync_val_compare_and_swap_16(__int128 *ptr, __int128 oldval, __int128 newval) {
  typedef union {
    struct parts {
      long lo; long hi;
    } parts;
    __int128 all;
  } k;

  k old_bits, new_bits;
  long temp_lo, temp_hi;
  old_bits.all = oldval;
  new_bits.all = newval;

  int result = 0;
  asm("1:\n"
      "ldaxp %[temp_lo], %[temp_hi], [%[ptr]]\n"
      "cmp %[temp_lo], %[old_lo]\n"
      "b.ne 0f\n"
      "cmp %[temp_hi], %[old_hi]\n"
      "b.ne 0f\n"
      "stlxp %w[result], %[new_lo], %[new_hi], [%[ptr]]\n"
      "cbnz %w[result], 1b\n"
      "mov %[result], #1\n"
      "0:\n"
      : [result]"+r"(result), [temp_lo]"=&r"(temp_lo), [temp_hi]"=&r"(temp_hi)
      : [ptr]"r"(ptr),
        [old_lo]"r"(old_bits.parts.lo), [old_hi]"r"(old_bits.parts.hi),
        [new_lo]"r"(new_bits.parts.lo), [new_hi]"r"(new_bits.parts.hi)
      : "memory");

  return result;
}

-- 
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671