Disable asan memory leak checks on a Cuda test. #9098
Conversation
Codecov Report
@@ Coverage Diff @@
## master #9098 +/- ##
==========================================
+ Coverage 57.03% 58.51% +1.47%
==========================================
Files 188 188
Lines 42211 42211
Branches 8766 8766
==========================================
+ Hits 24077 24698 +621
+ Misses 15908 15224 -684
- Partials 2226 2289 +63
Continue to review full report at Codecov.
|
|
@obilaniu since you've touched the cuda module a lot, perhaps you know why this started to fail to begin with? |
|
@eli-schwartz Hard to say! I did touch #8396 concerning the sanitizers, but that was CUDA test case 15. This is test case 16 concerning support for multiple CUDA standards, reported #8523 and fixed in #8527 / #8596. That was 3 months ago. At the time, I seem to recall the CUDA Arch runner was using GCC 10 and CUDA 11.2; Now I see it is using GCC 11 and CUDA 11.3. Perhaps mixed C++ standards and sanitizers play badly, or the new GCC or CUDA Toolkit expose a problem? |
|
When I was trying to figure out why it failed, I tried running multiple cuda test cases locally using meson.py setup builddir. They all produced asan erroring executables, but only one test case actually executes the program in a testsuite. So it's either a general problem with the cuda toolkit itself, or a problem with all our example program sources. I really don't think it's about any PR specifically. I was hoping maybe you knew some answer like "yeah, current versions of cuda are just buggy". |
|
@eli-schwartz It looks like |
Should fix Arch Cuda CI failure.