Constants in bli_const.c "undefined" on OSX about blis

[devinamatthews]

Two updates:

1. Both methods mentioned above seem to fix the problem.
2. The gemm_cntl* and trsm_[lr]_cntl* constants and possibly others are also affected.

[fgvanzee]

1. This is super annoying. I wonder if it runs afoul of any standards.
2. I suppose the most maintainable way of addressing this is solution 1. But you are right, it would affect the control tree as well. Can we #define an initializer expression in the spirit of PTHREAD_MUTEX_INITIALIZER?

[devinamatthews]

1. Not a standards issue, just a Mach limitation.
2. I am actually leaning towards 1. since it is a global change. If you do want to initialize, though, default-initialization with {} will just fill it with zeros instead of random data which should be fine.

[fgvanzee]

Sure, {} is fine for now. I may go back and replace them with _INITIALIZER macros in the future.

[devinamatthews]

I'll set up a PR.

On May 9, 2016 4:37:39 PM CDT, "Field G. Van Zee" [email protected] wrote:

Sure, {} is fine for now. I may go back and replace them with
_INITIALIZER macros in the future.

---

You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub:
#73 (comment)