Adding NUMBA_ENABLE_PROFILING envvar, enabling jit event#3307
Conversation
|
On Azure, the build error is not related to this PR: |
|
Yeah, this is a random failure that just started happening due to something strange (and apparently random) in the environment setup. It's almost like some of the build machines on Azure are behaving differently than others. |
|
Ok, I'll try to restart testing by force-pushing this commit |
|
Don't worry too much about it for now. We're treating the Azure results as "informative" while we evaluate how stable the service is. |
…fault value for NUMBA_DEBUGINFO
7fada39 to
ffe2bf8
Compare
|
@seibert could you please review the envvar name and changes to the defaults in this PR? I could sketch a doc section as well, if it works for you this way |
Codecov Report
@@ Coverage Diff @@
## master #3307 +/- ##
=========================================
Coverage ? 80.58%
=========================================
Files ? 390
Lines ? 78335
Branches ? 8854
=========================================
Hits ? 63126
Misses ? 13811
Partials ? 1398 |
|
This looks fine to me, but what code is calling |
|
Oops.. looks like it is a branch collision - I forgot I have this PR open and created this function for solving certain issues with multiple openmp copies. I'll roll it back |
73b25cb to
753282b
Compare
stuartarchibald
left a comment
There was a problem hiding this comment.
Thanks for the patch. On inspection this looks good. Will be interesting to see if perf can also be enabled in LLVM 7.
6fa4e08 to
c71d4c1
Compare
|
adding reference to numba/llvmlite#394 |
|
Testing this PR needs merge of #3424, then to do a merge build with master so as to pick up llvmlite 0.26 (where the linked changes are proposed). |
|
Smoke test build passed, only failures were unrelated. Merging! |
|
Thanks! |
|
No problem. Thanks for the contribution! |
Default value depends on whether profiling was detected
It also serves as a default value for NUMBA_DEBUGINFO
@sklam @seibert