orpiske commented on a change in pull request #5: ARTEMIS-2533 Support multiple
architecture for ASYNCIO kernel by-pass
URL:
https://github.com/apache/activemq-artemis-native/pull/5#discussion_r340316928
##########
File path:
src/main/c/org_apache_activemq_artemis_nativo_jlibaio_LibaioContext.c
##########
@@ -32,15 +32,11 @@
#include <fcntl.h>
#include <stdlib.h>
#include <pthread.h>
+#include <stdatomic.h>
#include <limits.h>
#include "org_apache_activemq_artemis_nativo_jlibaio_LibaioContext.h"
#include "exception_helper.h"
-//x86 has a strong memory model and there is no need of HW fences if just
Write-Back (WB) memory is used
-#define mem_barrier() __asm__ __volatile__ ("":::"memory")
-#define read_barrier() __asm__ __volatile__("":::"memory")
-#define store_barrier() __asm__ __volatile__("":::"memory")
-
Review comment:
Exactly. atomic_thread_fence is provided by stdatomic which, in turn, is
only available on newer GCC. To work-around that, @franz1981 can probably do
some thing like this:
1. Add a check for stdatomic.h on the CMakeLists.txt and, based on its
presence, define a HAVE_STDATOMIC_H flag:
```
include (CheckIncludeFiles)
if (UNIX)
CHECK_INCLUDE_FILES(stdatomic.h HAVE_STDATOMIC_H)
endif (UNIX)
```
2. Define atomic_thread_fence as the previous memory barrier to retain the
current behaviour:
```
#if defined(HAVE_STDATOMIC_H)
#include <stdatomic.h>
#else
#define memory_order_acquire ""
#define memory_order_release ""
#define atomic_thread_fence(x) __asm__ __volatile__ ("":::"memory")
#endif
```
This should make the code build on older versions just like the current one.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
