fiddle: release GVL for ffi_call

Some external functions I wish to call may take a long time
and unnecessarily block other threads.  This may lead to performance
regressions for fast functions as releasing/acquiring the GVL is not
cheap, but can improve performance for long-running functions
in multi-threaded applications.

This also means we must reacquire the GVL when calling Ruby-defined
callbacks for Fiddle::Closure, meaning we must detect whether the
current thread has the GVL by exporting ruby_thread_has_gvl_p
in internal.h

* ext/fiddle/function.c (struct nogvl_ffi_call_args):
  new struct for GVL release
  (nogvl_ffi_call): new function
  (function_call): adjust for GVL release
  [ruby-core:71642] [Feature #11607]
* ext/fiddle/closure.c (struct callback_args):
  new struct for GVL acquire
  (with_gvl_callback): adjusted original callback function
  (callback): wrapper for conditional GVL acquire
* ext/fiddle/depend: add dependencies
* ext/fiddle/extconf.rb: include top_srcdir for internal.h
* internal.h (ruby_thread_has_gvl_p): expose for fiddle
* vm_core.h (ruby_thread_has_gvl_p): moved to internal.h
* test/fiddle/test_function.rb (test_nogvl_poll): new test

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52723 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

1 files changed, 0 insertions, 1 deletions

diff --git a/vm_core.h b/vm_core.h
index aacdd3b897..da1455266d 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -1036,7 +1036,6 @@ rb_vm_living_threads_remove(rb_vm_t *vm, rb_thread_t *th)
     vm->living_thread_num--;
 }
 
-int ruby_thread_has_gvl_p(void);
 typedef int rb_backtrace_iter_func(void *, VALUE, int, VALUE);
 rb_control_frame_t *rb_vm_get_ruby_level_next_cfp(const rb_thread_t *th, const rb_control_frame_t *cfp);
 rb_control_frame_t *rb_vm_get_binding_creatable_next_cfp(const rb_thread_t *th, const rb_control_frame_t *cfp);