This reverts commit d3b6f835d565ec1590059773fc87589ddf8adc37.
This broke the Docker builds and presumably also 32-bit machines that
don't already have libatomic installed.
This was failing on crossruby, likely because HAVE_GCC_ATOMIC_BUILTINS
was true, but HAVE_GCC_ATOMIC_BUILTINS_64 was false. We probably should
have feature detection of 64-bit stdatomics like we do for GCC, but for
now let's keep rbimpl_atomic_u64_fetch_add in sync with load/set.
Previously this held a pointer to the Fiber itself, which requires
marking it (which was only implemented recently, prior to that it was
buggy). Using a monotonically increasing integer instead allows us to
avoid having a free function and keeps everything simpler.
My main motivations in making this change are that the root fiber lazily
allocates self, which makes the writebarrier implementation challenging
to do correctly, and wanting to avoid sending Mutexes to the remembered
set when locked by a short-lived Fiber.
This only adds the rbimpl_ version to include/ruby/atomic.h so that it
is not a new public interface.
We were already using RUBY_ATOMIC_VALUE_LOAD in a few locations. This
will allow us to use other memory orders internally when desired.
May not be supported on some 32bit architectures.
```
/usr/lib/gcc-cross/m68k-linux-gnu/14/../../../../m68k-linux-gnu/bin/ld: ../../libruby-static.a(vm.o): in function `rbimpl_atomic_u64_set_relaxed':
/home/ubuntu/build/ruby/master/m68k-linux/../src/ruby_atomic.h:60:(.text+0x2468): undefined reference to `__atomic_store_8'
/usr/lib/gcc-cross/m68k-linux-gnu/14/../../../../m68k-linux-gnu/bin/ld: ../../libruby-static.a(vm.o): in function `rbimpl_atomic_u64_load_relaxed':
/home/ubuntu/build/ruby/master/m68k-linux/../src/ruby_atomic.h:43:(.text+0x2950):
undefined reference to `__atomic_load_8'
```
Followup: https://github.com/ruby/ruby/pull/13341 / [Feature #21353]
Even thought `shape_id_t` has been make 32bits, we were still limited
to use only the lower 16 bits because they had to fit alongside `attr_index_t`
inside a `uintptr_t` in inline caches.
By enlarging inline caches we can unlock the full 32bits on all
platforms, allowing to use these extra bits for tagging.
There are cases when sizeof(int) == sizeof(long) == sizeof(size_t). On
such cases however int and long are incompatible types in theory. We
should not assume typedef long size_t, because on Solaris size_t is
actually a typedef of int.
This reduces compiler warnings on such situations.
With these macros implemented we can write codes just like we can assume
the compiler being clang. MSC_VERSION_SINCE is defined to implement
those macros, but turned out to be handy for other places. The -fdeclspec
compiler flag is necessary for clang to properly handle __has_declspec().
This macro shall be shared among other files, not just for mjit.c.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64156 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_CAS): old value to be swapped should be
same as the destination. immediate value may need type
promotion.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52893 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_SIZE_CAS): fix the argument order of
InterlockedCompareExchange64. new value and then old value is
the last.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52892 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_VALUE_CAS): fix typo.
TODO: make arguments of all CAS macros consistent.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51781 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_PTR_CAS): define by generic CAS macro, not
via size_t, to suppress a warning by mingw gcc.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@50984 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* iseq.c (iseq_load): keep type_map to get rid of memory leak.
based on a patch by Eric Wong at [ruby-core:59699]. [Bug #9399]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@44568 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_PTR_EXCHANGE): atomic exchange function for
a generic pointer.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@43996 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
__sync family are legacy functions now and it is recommended
that new code use the __atomic functions.
http://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html
* configure.in: check existence of __atomic functions.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@43851 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_SIZE_SUB): add cast to suppress a warning by VC++.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@43356 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h (ATOMIC_CAS): suppress C4022 and C4047 warnings in
VC6. only InterlockedCompareExchange is declared using PVOID.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@41759 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h: Skip using Solaris10 atomic_ops on Solaris 9 or earlier if atomic.h is not available. [ruby-dev:47229] [Bug #8228]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@40163 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* ruby_atomic.h, signal.c (NEED_RUBY_ATOMIC_OPS): renamed from
NEED_RUBY_ATOMIC_EXCHANGE.
* signal.c (ruby_atomic_compare_and_swap): naive, non-atomic
compare-and-swap implementation only used for platforms without
valid support for atomic operations.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@37897 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
