| commit | 6ce464aed89de37c519fad36d17fa9b9ba7819e5 | |
| author | Chris Robinson <chris.kcat@gmail.com> | |
| Thu, 31 Jul 2014 11:19:33 +0000 (31 04:19 -0700) | ||
| committer | Chris Robinson <chris.kcat@gmail.com> | |
| Thu, 31 Jul 2014 11:19:33 +0000 (31 04:19 -0700) | ||
| tree | 9ce2136ff4d1e2afd00b1792e6a4acc1e7a187c0 | treesnapshot (tar.gz zip) |
| parent | 7c847e62c3825621dc6ce2acc793e821d61ce057 | commitdiff |
Avoid a CPU memory fence, and better order memory barriers to accesses
Code produced by GCC 4.9's C11 atomic implementation does not seem to add a
CPU memory fence even with memory_order_seq_cst. Unnecessary memory fences
will be a detriment to performance, so they should only be added if needed.
Code produced by GCC 4.9's C11 atomic implementation does not seem to add a
CPU memory fence even with memory_order_seq_cst. Unnecessary memory fences
will be a detriment to performance, so they should only be added if needed.
| include/atomic.h | diffblobblamehistory |