[Mesa-dev] [PATCH] configure.ac: rework -latomic check

[Matt Turner]

On Thu, May 10, 2018 at 1:43 AM, Thomas Petazzoni
<thomas.petazzoni at bootlin.com> wrote:
> Hello Matt,
>
> On Wed,  9 May 2018 16:30:12 -0700, Matt Turner wrote:
>
>> Hi Thomas,
>>
>> I rebased this patch on
>>
>>       commit 54ba73ef102f7b9085922686bb31719539e0dc3c
>>       Author: Nicolas Boichat <drinkcat at chromium.org>
>>       Date:   Thu Apr 5 09:33:09 2018 +0800
>>
>>           configure.ac/meson.build: Fix -latomic test
>>
>> and readded the AC_MSG_CHECKING/AC_MSG_RESULT. If it looks good to you,
>> I'll commit it.
>
> Thanks for this new iteration! I didn't test it, but it looks good to
> me. Just one minor nit below.
>
>> -dnl Check for new-style atomic builtins
>> -AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
>> +dnl Check for new-style atomic builtins. We first check without linking to
>> +dnl -latomic.
>> +AC_MSG_CHECKING(whether __atomic_load_n is supported)
>> +AC_LINK_IFELSE([AC_LANG_SOURCE([[
>> +#include <stdint.h>
>>  int main() {
>> -    int n;
>> -    return __atomic_load_n(&n, __ATOMIC_ACQUIRE);
>> -}]])], GCC_ATOMIC_BUILTINS_SUPPORTED=1)
>> -if test "x$GCC_ATOMIC_BUILTINS_SUPPORTED" = x1; then
>> -    DEFINES="$DEFINES -DUSE_GCC_ATOMIC_BUILTINS"
>> -    dnl On some platforms, new-style atomics need a helper library
>> -    AC_MSG_CHECKING(whether -latomic is needed)
>> -    AC_LINK_IFELSE([AC_LANG_SOURCE([[
>> -    #include <stdint.h>
>>      struct {
>>          uint64_t *v;
>>      } x;
>> -    int main() {
>> +    return (int)__atomic_load_n(x.v, __ATOMIC_ACQUIRE);
>> +}]])], GCC_ATOMIC_BUILTINS_SUPPORTED=yes, GCC_ATOMIC_BUILTINS_SUPPORTED=no)
>> +
>> +dnl If that didn't work, we try linking with -latomic, which is needed on some
>> +dnl platforms.
>> +if test "x$GCC_ATOMIC_BUILTINS_SUPPORTED" != xyes; then
>> +   save_LDFLAGS=$LDFLAGS
>> +   LDFLAGS="$LDFLAGS -latomic"
>> +   AC_LINK_IFELSE([AC_LANG_SOURCE([[
>> +   #include <stdint.h>
>> +   int main() {
>> +        struct {
>> +            uint64_t *v;
>> +        } x;
>>          return (int)__atomic_load_n(x.v, __ATOMIC_ACQUIRE);
>> -    }]])], GCC_ATOMIC_BUILTINS_NEED_LIBATOMIC=no, GCC_ATOMIC_BUILTINS_NEED_LIBATOMIC=yes)
>> -    AC_MSG_RESULT($GCC_ATOMIC_BUILTINS_NEED_LIBATOMIC)
>> -    if test "x$GCC_ATOMIC_BUILTINS_NEED_LIBATOMIC" = xyes; then
>> -        LIBATOMIC_LIBS="-latomic"
>> -    fi
>> +   }]])], GCC_ATOMIC_BUILTINS_SUPPORTED=yes LIBATOMIC_LIBS="-latomic",
>> +          GCC_ATOMIC_BUILTINS_SUPPORTED=no)
>> +   LDFLAGS=$save_LDFLAGS
>> +
>> +
>
> Those two empty lines don't seem to be very useful.

Thanks for catching that. I think I was toying with another
AC_MSG_CHECKING for whether -latomic is needed and decided against it.

Fixed and pushed. Thanks for the patch!